HPC Compass IBM Special 2013/14

Simulation
Automotive
CAD
CAEEngineering
Price Modelling
Life Sciences
Aerospace
Risk Analysis
Big Data Analytics
High Throughput Computing
HighPerformance
Computing2013/14
TechnologyCompass
IBMSpecial

2 3
More than 30 years of experience in
Scientific Computing
1980 marked the beginning of a decade where numerous startups
were created, some of which later transformed into big players in
the IT market. Technical innovations brought dramatic changes to
the nascent computer market. In Tübingen, close to one of Germa-
ny’s prime and oldest universities, transtec was founded.
In the early days, transtec focused on reselling DEC computers and
peripherals, delivering high-performance workstations to univer-
sity institutes and research facilities. In 1987, SUN/Sparc and stor-
age solutions broadened the portfolio, enhanced by IBM/RS6000
products in 1991. These were the typical workstations and server
systems for high performance computing then, used by the major-
ity of researchers worldwide.
In the late 90s, transtec was one of the first companies to offer
highly customized HPC cluster solutions based on standard Intel
architecture servers, some of which entered the TOP500 list of the
world’s fastest computing systems.
Thus, given this background and history, it is fair to say that trans-
tec looks back upon a more than 30 years’ experience in scientific
computing; our track record shows nearly 500 HPC installations.
With this experience, we know exactly what customers’ demands
are and how to meet them. High performance and ease of manage-
ment – this is what customers require today. HPC systems are for
sure required to peak-perform, as their name indicates, but that is
not enough: they must also be easy to handle. Unwieldy design and
operationalcomplexitymustbeavoidedoratleasthiddenfromad-
ministrators and particularly users of HPC computer systems.
transtec HPC solutions deliver ease of management, both in the
Linux and Windows worlds, and even where the customer´s envi-
ronment is of a highly heterogeneous nature. Even the dynamical
provisioning of HPC resources as needed does not constitute any
problem, thus further leading to maximal utilization of the cluster.
transtec HPC solutions use the latest and most innovative technol-
ogy. Their superior performance goes hand in hand with energy ef-
ficiency, as you would expect from any leading edge IT solution. We
regard these basic characteristics.
In2010,transtecenteredintoastrategicpartnershipwithIBM,sure-
ly one of the biggest players in the HPC world with a very strong
brand. The flexibility and long-year experience of transtec, com-
bined with the power and quality of IBM HPC systems constitute
a perfect symbiosis and provide customers with the most optimal
HPC solution imaginable. IBM iDataPlex systems are highly opti-
mized for HPC workload in datacenter environments, regarding
performance, flexibility, and energy, space and cooling efficiency.
Platform HPC and LSF are both enterprise-ready HPC cluster and
workloadmanagementsolutionsandarewidespreadinallkindsof
industrial HPC environments.
Your decision for a transtec HPC solution means you opt for most
intensive customer care and best service in HPC. Our experts will
be glad to bring in their expertise and support to assist you at any
stage, from HPC design to daily cluster operations, to HPC Cloud
Services.
Last but not least, transtec HPC Cloud Services provide customers
with the possibility to have their jobs run on dynamically provided
nodes in a dedicated datacenter, professionally managed and indi-
viduallycustomizable.NumerousstandardapplicationslikeANSYS,
LS-Dyna, OpenFOAM, as well as lots of codes like Gromacs, NAMD,
VMD, and others are pre-installed, integrated into an enterprise-
ready cloud management environment, and ready to run.
Have fun reading the transtec HPC Compass 2013/14 IBM Special!
TechnologyCompass
Table of contents and Introduction
Enterprise-Ready Cluster &
Workload Management...........................................................4
IBM Platform HPC..................................................................................................6
What’s New in IBM Platform LSF 8 ..........................................................10
IBM Platform MPI 8.1........................................................................................20
High Performance Computing......................................... 24
Performance Turns Into Productivity....................................................24
Scalable & Energy Efficient HPC Systems.................. 26
IBM iDataPlex dx360 m4 HPC Clusters ..................................................28
General Parallel File System (GPFS)............................... 36
Technologies That Enable the Management of Big Data..........38
What´s New in GPFS Version 3.5...............................................................54

5
Enterprise-Ready
Cluster&Workload
Management
High performance comput-
ing (HPC) is becoming a nec-
essary tool for organizations
to speed up product design,
scientific research, and busi-
ness analytics. However,
there are few software en-
vironments more complex
to manage and utilize than
modern high performance
computing clusters. There-
fore, addressing the prob-
lem of complexity in cluster
management is a key aspect
ofleveragingHPCtoimprove
time to results and user pro-
ductivity.

Application-Centric Interface
Uniﬁed Management Interface
Provisioning
& Node
Management
Workload
Management
Parallel Job
Enablement
Adaptive
Scheduling
Essential components of an HPC cluster solution
6 7
Introduction
Clusters based on the Linux® operating system have become
increasingly prevalent at large supercomputing centers and
continue to make significant in-roads in commercial and aca-
demic settings. This is primarily due to their superior price/per-
formance and flexibility, as well as the availability of commercial
applications that are based on the Linux OS.
Ironically, the same factors that make Linux a clear choice for
high performance computing often make the operating system
less accessible to smaller computing centers. These organiza-
tions may have Microsoft Windows administrators on staff, but
have little or no Linux or cluster management experience. The
complexity and cost of cluster management often outweigh the
benefits that make open, commodity clusters so compelling. Not
only can HPC cluster deployments be difficult, but the ongoing
need to deal with heterogeneous hardware and operating sys-
tems, mixed workloads, and rapidly evolving toolsets make de-
ploying and managing an HPC cluster a daunting task.
These issues create a barrier to entry for scientists and research-
ers who require the performance of an HPC cluster, but are lim-
ited to the performance of a workstation. This is why ease of use
is now mandatory for HPC cluster management. This paper re-
views the most complete and easy to use cluster management
solution, Platform HPC, which is now commercially available
from Platform Computing.
The cluster management challenge
To provide a proper HPC application environment, system admin-
istrators need to provide a full set of capabilities to their users,
as shown below. These capabilities include cluster provisioning
and node management, application workload management, and
Enterprise-ReadyCluster&
WorkloadManagement
IBM Platform HPC
an environment that makes it easy to develop, run and manage
distributed parallel applications.
Modern application environments tend to be heterogeneous;
some workloads require Windows compute hosts while oth-
ers require particular Linux operating systems or versions.
The ability to change a node’s operating system on-the-fly in re-
sponse to changing application needs - referred to as adaptive
scheduling - is important since it allows system administrators
to maximize resource use, and present what appears to be a larg-
er resource pool to cluster users.
Learning how to use a command line interface to power-up, pro-
vision and manage a cluster is extremely time-consuming. Ad-
ministrators therefore need remote, web-based access to their
HPC environment that makes it easier for them to install and
manage an HPC cluster. An easy-to-use application-centric web
interface can have tangible benefits including improved produc-
tivity, reduced training requirements, reduced errors rates, and
secure remote access.
While there are several cluster management tools that address
parts of these requirements, few address them fully, and some
tools are little more than collections of discrete open-source
software components.
Some cluster toolkits focus largely on the problem of cluster pro-
visioning and management. While they clearly simplify cluster
deployment, administrators wanting to make changes to node
configurations or customize their environment will quickly
find themselves hand-editing XML configuration files or writ-
ing their own shell scripts. Third-party workload managers and
various open-source MPI libraries might be included as part of
a distribution. However, these included components are loosely
integrated and often need to be managed separately from the
cluster manager. As a result the cluster administrator needs to
learn how to utilize each additional piece of software in order to
manage the cluster effectively.
Other HPC solutions are designed purely for application work-
load management. While these are all capable workload manag-
ers, most do not address at all the issue of cluster management,
application integration, or adaptive scheduling. If such capabili-
ties exist they usually require the purchase of additional soft-
ware products.
Parallel job management is also critical. One of the primary rea-
sons that customers deploy HPC clusters is to maximize applica-
tion performance. Processing problems in parallel is a common
way to achieve performance gains. The choice of MPI, its scalabil-
ity, and the degree to which it is integrated with various OFED
drivers and high performance interconnects has a direct impact
on delivered application performance. Furthermore, if the work-
load manager does not incorporate specific parallel job manage-
ment features, busy cluster users and administrators can find
themselves manually cleaning up after failed MPI jobs or writing
their own shell scripts to do the same.

“Platform HPC and Platform LSF have been known
for many years as a highest-quality and enterprise-
ready cluster and workload management solutions,
and we have many customers in academia and in-
dustry relying on them.”
Dr. Oliver Tennert Director Technology Management & HPC Solutions
8 9
WorkloadManagement
IBM Platform HPC
Complexity is a real problem. Many small organizations or de-
partments grapple with a new vocabulary full of cryptic com-
mands, configuring and troubleshooting Anaconda kick start
scripts, finding the correct OFED drivers for specialized hard-
ware, and configuring open source monitoring systems like Gan-
glia or Nagios. Without an integrated solution administrators
may need to deal with dozens of distinct software components,
making managing HPC cluster implementations extremely te-
dious and time-consuming.
Re-thinking HPC clusters
Clearly these challenges demand a fresh approach to HPC clus-
ter management. Platform HPC represents a “re-think” of how
HPC clusters are deployed and managed. Rather than address-
ing only part of the HPC management puzzle, Platform HPC ad-
dresses all facets of cluster management. It provides:
ʎʎ A complete, easy-to-use cluster management solution
ʎʎ Integrated application support
ʎʎ User-friendly, topology-aware workload management
ʎʎ Robust workload and system monitoring and reporting
ʎʎ Dynamic operating system multi-boot (adaptive scheduling)
ʎʎ GPU scheduling
ʎʎ Robust commercial MPI library (Platform MPI)
ʎʎ Web-based interface for access anywhere
Most complete HPC cluster management solution
PlatformHPCmakesiteasytodeploy,runandmanageHPCclusters
while meeting the most demanding requirements for application
performance and predictable workload management. It is a com-
plete solution that provides a robust set of cluster management ca-
pabilities; from cluster provisioning and management to workload
management and monitoring. The easy-to -use unified web portal
provides a single point of access into the cluster, making it easy to
manage your jobs and optimize application performance.
Platform HPC is more than just a stack of software; it is a fully in-
tegrated and certified solution designed to ensure ease of use and
simplified troubleshooting.
Integrated application support
High performing, HPC-optimized MPI libraries come integrated
with Platform HPC, making it easy to get parallel applications up
and running. Scripting guidelines and job submission templates for
commonly used commercial applications simplify job submission,
reducesetuptimeandminimizeoperationerrors.Oncetheapplica-
tions are up and running, Platform HPC improves application per-
formance by intelligently scheduling resources based on workload
characteristics.
Fully certified and supported
Platform HPC unlocks cluster management to provide the easiest
and most complete HPC management capabilities while reducing
overall cluster cost and improving administrator productivity. It is
based on the industry’s most mature and robust workload manag-
er, Platform LSF, making it the most reliable solution on the market.
Other solutions are typically a collection of open-source tools,
which may also include pieces of commercially developed soft-
ware. They lack key HPC functionality and vendor support, relying
on the administrator’s technical ability and time to implement.
Platform HPC is a single product with a single installer and a uni-
fiedweb-basedmanagementinterface.Withthebestsupportinthe
HPC industry, Platform HPC provides the most complete solution
for HPC cluster management.
Complete solution
Platform HPC provides a complete set of HPC cluster manage-
ment features. In this section we’ll explore some of these unique
capabilities in more detail.
Easy to use, cluster provisioning and management
With Platform HPC, administrators can quickly provision and
manage HPC clusters with unprecedented ease. It ensures maxi-
mum uptime and can transparently synchronize files to cluster
nodes without any downtime or re-installation.
Fast and efficient software Installation – Platform HPC can be
installed on the head node and takes less than one hour using
three different mechanisms:
ʎʎ Platform HPC DVD
ʎʎ Platform HPC ISO file
ʎʎ Platform partner’s factory install bootable USB drive
Installing software on cluster nodes is simply a matter of associ-
ating cluster nodes with flexible provisioning templates through
the web-based interface.
Flexible provisioning – Platform HPC offers multiple options for
provisioning Linux operating environments that include:
ʎʎ Package-based provisioning
ʎʎ Image based provisioning
ʎʎ Diskless node provisioning
Large collections of hosts can be provisioned using the same pro-
visioning template. Platform HPC automatically manages details
such as IP address assignment and node naming conventions that
reflect the position of cluster nodes in data center racks.

10 11
What’s New in IBM Platform LSF 8
Written with Platform LSF administrators in mind, this brief pro-
vides a short explanation of significant changes in Platform’s lat-
est release of Platform LSF, with a specific emphasis on schedul-
ing and workload management features.
About IBM Platform LSF 8
Platform LSF is the most powerful workload manager for de-
manding, distributed high performance computing environ-
ments. It provides a complete set of workload management
capabilities, all designed to work together to reduce cycle times
and maximize productivity in missioncritical environments.
This latest Platform LSF release delivers improvements in perfor-
mance and scalability while introducing new features that sim-
plify administration and boost user productivity. This includes:
ʎʎ Guaranteed resources – Aligns business SLA’s with infra-
structure configuration for simplified administration and
configuration
ʎʎ Live reconfiguration – Provides simplified administration
and enables agility
ʎʎ Delegation of administrative rights – Empowers line of busi-
ness owners to take control of their own projects
ʎʎ Fairshare & pre-emptive scheduling enhancements – Fine
tunes key production policies
Platform LSF 8 Features
Guaranteed Resources Ensure Deadlines are Met
In Platform LSF 8, resource-based scheduling has been extended
to guarantee resource availability to groups of jobs. Resources
can be slots, entire hosts or user-defined shared resources such
as software licenses. As an example, a business unit might guar-
antee that it has access to specific types of resources within ten
minutes of a job being submitted, even while sharing resources
between departments. This facility ensures that lower priority
jobs using the needed resources can be pre-empted in order to
meet the SLAs of higher priority jobs.
Because jobs can be automatically attached to an SLA class via
access controls, administrators can enable these guarantees
without requiring that end-users change their job submission
procedures, making it easy to implement this capability in exist-
ing environments.
Live Cluster Reconfiguration
Platform LSF 8 incorporates a new live reconfiguration capabil-
ity, allowing changes to be made to clusters without the need
to re-start LSF daemons. This is useful to customers who need
to add hosts, adjust sharing policies or re-assign users between
groups “on the fly”, without impacting cluster availability or run-
ning jobs.
Changes to the cluster configuration can be made via the bconf
command line utility, or via new API calls. This functionality can
also be integrated via a web-based interface using Platform Ap-
plication Center. All configuration modifications are logged for
a complete audit history, and changes are propagated almost
instantaneously. The majority of reconfiguration operations are
completed in under half a second.
Unlike competing solutions, Platform HPC deploys multiple op-
erating systems and OS versions to a cluster simultaneously. This
includes Red Hat® Enterprise Linux, CentOS, Scientific Linux, and
SUSE® Linux Enterprise Server. This provides administrators with
greater flexibility in how they serve their user communities and
means that HPC clusters can grow and evolve incrementally as re-
quirements change.
WorkloadManagement

1312
With Live Reconfiguration, down-time is reduced, and administra-
torsarefreetomakeneededadjustmentsquicklyratherthanwait
for scheduled maintenance periods or non-peak hours. In cases
where users are members of multiple groups, controls can be put
in place so that a group administrator can only control jobs as-
sociated with their designated group rather than impacting jobs
related to another group submitted by the same user.
Delegation of Administrative Rights
With Platform LSF 8, the concept of group administrators has been
extended to enable project managers and line of business man-
agers to dynamically modify group membership and fairshare re-
source allocation policies within their group. The ability to make
these changes dynamically to a running cluster is made possible by
the Live Reconfiguration feature.
These capabilities can be delegated selectively depending on the
group and site policy. Different group administrators can manage
jobs, control sharing policies or adjust group membership.
More Flexible Fairshare Scheduling Policies
ToenablebetterresourcesharingflexibilitywithPlatformLSF8,the
algorithms used to tune dynamically calculated user priorities can
be adjusted at the queue level. These algorithms can vary based on
department, application or project team preferences. The Fairshare
parameters ENABLE_HIST_RUN_TIME and HIST_HOURS enable
administrators to control the degree to which LSF considers prior
resource usage when determining user priority. The flexibility of
Platform LSF 8 has also been improved by allowing a similar “decay
rate” to apply to currently running jobs (RUN_TIME_DECAY), either
system-wide or at the queue level. This is most useful for custom-
ers with long-running jobs, where setting this parameter results in
a more accurate view of real resource use for the fairshare schedul-
ing to consider.
Performance & Scalability Enhancements
Platform LSF has been extended to support an unparalleled
scale of up to 100,000 cores and 1.5 million queued jobs for
very high throughput EDA workloads. Even higher scalability
is possible for more traditional HPC workloads.
Specific areas of improvement include the time required to
start the master-batch daemon (MBD), bjobs query perfor-
mance, job submission and job dispatching as well as impres-
sive performance gains resulting from the new Bulk Job Sub-
mission feature. In addition, on very large clusters with large
numbers of user groups employing fairshare scheduling, the
memory footprint of the master batch scheduler in LSF has
been reduced by approximately 70% and scheduler cycle time
has been reduced by 25%, resulting in better performance and
scalability.
More Sophisticated Host-based Resource Usage for Parallel Jobs
Platform LSF 8 provides several improvements to how resource
use is tracked and reported with parallel jobs. Accurate tracking
of how parallel jobs use resources such as CPUs, memory and
swap, is important for ease of management, optimal scheduling
and accurate reporting and workload analysis. With Platform
LSF 8 administrators can track resource usage on a per-host
basis and an aggregated basis (across all hosts), ensuring that
resource use is reported accurately. Additional details such as
running PIDs and PGIDs for distributed parallel jobs, manual
cleanup (if necessary) and the development of scripts for manag-
ing parallel jobs are simplified. These improvements in resource
usage reporting are reflected in LSF commands including bjobs,
bhist and bacct.
Improved Ease of Administration for Mixed Windows and
Linux Clusters
The lspasswd command in Platform LSF enables Windows LSF
users to advise LSF of changes to their Windows level pass-
words. With Platform LSF 8, password synchronization between
environments has become much easier to manage because the
Windows passwords can now be adjusted directly from Linux
hosts using the lspasswd command. This allows Linux users to
conveniently synchronize passwords on Windows hosts without
needing to explicitly login into the host.
Bulk Job Submission
When submitting large numbers of jobs with different resource
requirements or job level settings, Bulk Job Submission allows
for jobs to be submitted in bulk by referencing a single file con-
taining job details.
WorkloadManagement

Resource monitoring
14 15
Simplified configuration changes – Platform HPC simplifies ad-
ministration and increases cluster availability by allowing chang-
es such as new package installations, patch updates, and changes
to configuration files to be propagated to cluster nodes automati-
cally without the need to re-install those nodes. It also provides a
mechanism whereby experienced administrators can quickly per-
form operations in parallel across multiple cluster nodes.
Repository snapshots / trial installations – Upgrading software
can be risky, particularly in complex environments. If a new soft-
ware upgrade introduces problems, administrators often need
to rapidly “rollback” to a known good state. With other cluster
managers this can mean having to re-install the entire cluster.
Platform HPC incorporates repository snapshots, which are “re-
store points” for the entire cluster. Administrators can snapshot a
known good repository, make changes to their environment, and
easily revert to a previous “known good” repository in the event of
anunforeseenproblem.Thispowerfulcapabilitytakestheriskout
of cluster software upgrades.
New hardware integration – When new hardware is added
to a cluster it may require new or updated device drivers that
are not supported by the OS environment on the installer
node. This means that a newly updated node may not net-
work boot and provision until the head node on the cluster is
updated with a new operating system; a tedious and disrup-
tive process. Platform HPC includes a driver patching utility
that allows updated device drivers to be inserted into exist-
ing repositories, essentially future proofing the cluster, and
providing a simplified means of supporting new hardware
without needing to re-install the environment from scratch.
Software updates with no re-boot – Some cluster managers al-
ways re-boot nodes when updating software, regardless of how
minor the change. This is a simple way to manage updates. How-
ever, scheduling downtime can be difficult and disruptive. Platform
HPCperformsupdatesintelligentlyandselectivelysothatcompute
nodes continue to run even as non-intrusive updates are applied.
The repository is automatically updated so that future installations
include the software update. Changes that require the re-installa-
tion of the node (e.g. upgrading an operating system) can be made
in a “pending” state until downtime can be scheduled.
User-friendly, topology aware workload manage-
ment
Platform HPC includes a robust workload scheduling capability,
which is based on Platform LSF - the industry’s most powerful,
comprehensive, policy driven workload management solution for
engineering and scientific distributed computing environments.
By scheduling workloads intelligently according to policy,
Platform HPC improves end user productivity with minimal sys-
tem administrative effort. In addition, it allows HPC user teams to
easily access and share all computing resources, while reducing
time between simulation iterations.
GPU scheduling – Platform HPC provides the capability to sched-
ule jobs to GPUs as well as CPUs. This is particularly advantageous
in heterogeneous hardware environments as it means that ad-
ministrators can configure Platform HPC so that only those jobs
that can benefit from running on GPUs are allocated to those re-
sources. This frees up CPU-based resources to run other jobs. Us-
ing the unified management interface, administrators can moni-
tor the GPU performance as well as detect ECC errors.
Unified management interface
Competing cluster management tools either do not have a web-
based interface or require multiple interfaces for managing dif-
ferent functional areas. In comparison, Platform HPC includes a
single unified interface through which all administrative tasks
can be performed including node-management, job-manage-
ment, jobs and cluster monitoring and reporting. Using the uni-
fied management interface, even cluster administrators with
very little Linux experience can competently manage a state of
the art HPC cluster.
Job management – While command line savvy users can contin-
ue using the remote terminal capability, the unified web portal
makes it easy to submit, monitor, and manage jobs. As changes
are made to the cluster configuration, Platform HPC automati-
cally re-configures key components, ensuring that jobs are allo-
cated to the appropriate resources. The web portal is customiz-
WorkloadManagement
IBM Platform HPC

Job submission templates
16 17
able and provides job data management, remote visualization
and interactive job support.
Workload/system correlation – Administrators can correlate
workload information with system load, so that they can make
timely decisions and proactively manage compute resources
against business demand. When it’s time for capacity planning,
the management interface can be used to run detailed reports
and analyses which quantify user needs and remove the guess
work from capacity expansion.
Simplified cluster management – The unified management con-
sole is used to administer all aspects of the cluster environment. It
enables administrators to easily install, manage and monitor their
cluster. It also provides an interactive environment to easily pack-
age software as kits for application deployment as well as pre-in-
tegrated commercial application support. One of the key features
of the interface is an operational dashboard that provides com-
prehensive administrative reports. As the image illustrates, Plat-
form HPC enables administrators to monitor and report on key
performance metrics such as cluster capacity, available memory
and CPU utilization. This enables administrators to easily identify
and troubleshoot issues.
The easy to use interface saves the cluster administrator time, and
means that they do not need to become an expert in the adminis-
tration of open-source software components. It also reduces the
possibility of errors and time lost due to incorrect configuration.
Cluster administrators enjoy the best of both worlds – easy ac-
cess to a powerful, web-based cluster manager without the need
to learn and separately administer all the tools that comprise the
HPC cluster environment.
Robust Commercial MPI library
Platform MPI – In order to make it easier to get parallel applica-
tions up and running, Platform HPC includes the industry’s most
robust and highest performing MPI implementation, Platform
MPI. Platform MPI provides consistent performance at application
run-time and for application scaling, resulting in top performance
results across a range of third-party benchmarks.
Open Source MPI – Platform HPC also includes various other in-
dustry standard MPI implementations. This includes MPICH1,
MPICH2 and MVAPICH1, which are optimized for cluster hosts con-
nected via InfiniBand, iWARP or other RDMA based interconnects.
Integrated application support
Job submission templates – Platform HPC comes complete with
job submission templates for ANSYS Mechanical, ANSYS Flu-
ent, ANSYS CFX, LS-DYNA, MSC Nastran, Schlumberger ECLIPSE,
Simulia Abaqus, NCBI Blast, NWChem, ClustalW, and HMMER. By
configuring these templates based on the application settings
in your environment, users can start using the cluster without
writing scripts.
Scripting Guidelines – Cluster users that utilize homegrown or
open-source applications, can utilize the Platform HPC scripting
guidelines. These user-friendly interfaces help minimize job sub-
mission errors. They are also self-documenting, enabling users to
create their own job submission templates.
Benchmark tests – Platform HPC also includes standard bench-
mark tests to ensure that your cluster will deliver the best per-
formance without manual tuning.
Flexible OS provisioning
Platform HPC can deploy multiple operating system versions con-
currently on the same cluster and, based on job resource require-
ments, dynamically boot the Linux or Windows operating system
required to run the job. Administrators can also use a web inter-
face to manually switch nodes to the required OS to meet applica-
tion demands, providing them with the flexibility to support spe-
cial requests and accommodate unanticipated changes. Rather
than being an extracost item as it is with other HPC management
suites,thiscapabilityisincludedasacorefeatureofPlatformHPC.
Commercial Service and support
Certified cluster configurations – Platform HPC is tested and
certified on all partner hardware platforms. By qualifying each
platform individually and providing vendor-specific software
with optimized libraries and drivers that take maximum ad-
vantage of unique hardware features, Platform Computing has
essentially done the integration work in advance. As a result,
WorkloadManagement
IBM Platform HPC

18 19
clusters can be deployed quickly and predictably with minimal
effort. As a testament to this, Platform HPC is certified under the
Intel Cluster Ready program.
Enterprise class service and support – Widely regarded as hav-
ing the best HPC support organization in the business, Platform
Computing is uniquely able to support an integrated HPC plat-
form. Because support personnel have direct access to the Plat-
form HPC developers, Platform Computing is able to offer a high-
er level of support and ensure that any problems encountered
are resolved quickly and efficiently.
Summary
Platform HPC is the ideal solution for deploying and manag-
ing state of the art HPC clusters. It makes cluster management
simple, enabling analysts, engineers and scientists from organi-
zations of any size to easily exploit the power of Linux clusters.
Unlike other HPC solutions that address only parts of the HPC
management challenge, Platform HPC uniquely addresses all as-
pects of cluster and management including:
ʎʎ Easy-to-use cluster provisioning and management
ʎʎ User-friendly, topology aware workload management
ʎʎ Unified management interface
ʎʎ Robust commercial MPI library
ʎʎ Integrated application support
ʎʎ Flexible OS provisioning
ʎʎ Commercial HPC service and support
By providing simplified management over the entire lifecycle of
a cluster, Platform HPC has a direct and positive impact on pro-
ductivity while helping to reduce complexity and cost. The com-
prehensive web-based management interface, and features like
repository snapshots and the ability to update software pack-
ages on the fly means that state-of-the-art HPC clusters can be
provisioned and managed even by administrators with little or
no Linux administration experience.
Capability / Feature Platform HPC
Cluster Provisioning and Management x
Initial cluster provisioning x
Multiple provisioning methods x
Web-based cluster management x
Node updates with no re-boot x
Repository snapshots x
Flexible node templates x
Multiple OS and OS versions x
Workload Management & Application Integration x
Integrated workload management x
HPC libraries & toolsets x
NVIDIA CUDA SDK support x
Web-based job management x
Web-based job data management x
Multi-boot based on workload x
Advanced parallel job management x
Commercial application integrations x
MPI Libraries x
Commercial grade MPI x
Workload and system monitoring, reporting and
correlation
x
Workload monitoring x
Workload reporting x
System monitoring & reporting x
Workload and system load correlation x
Integration with 3rd party management tools x
WorkloadManagement
IBM Platform HPC

20 21
IBM Platform MPI 8.1
Benefits
ʎʎ Superior application performance
ʎʎ Reduced development and support costs
ʎʎ Faster time-to-market
ʎʎ The industry’s best technical support
ʎʎ
Features
ʎʎ Supports the widest range of hardware, networks and oper-
ating systems
ʎʎ Distributed by over 30 leading commercial software vendors
ʎʎ Change interconnects or libraries with no need to re-compile
ʎʎ Seamless compatibility across Windows® and Linux®
environments
ʎʎ Ensures a production quality implementation
Ideal for:
ʎʎ Enterprises that develop or deploy parallelized software ap-
plications on HPC clusters
ʎʎ Commercial software vendors wanting to improve applica-
tions performance over the widest range of computer hard-
ware, interconnects and operating systems
The Standard for Scalable, Parallel Applications
Platform MPI is a high performance, production–quality imple-
mentation of the Message Passing Interface (MPI). It is widely
used in the high performance computing (HPC) industry and is
considered the de facto standard for developing scalable, paral-
lel applications.
Platform MPI maintains full backward compatibility with HP-MPI
and Platform MPI applications and incorporates advanced CPU
affinity features, dynamic selection of interface libraries, supe-
rior workload manger integrations and improved performance
and scalability.
Platform MPI supports the broadest range of industry standard
platforms, interconnects and operating systems helping ensure
that your parallel applications can run anywhere.
Focus on portability
Platform MPI allows developers to build a single executable that
transparently leverages the performance features of any type of
interconnect, thereby providing applications with optimal laten-
cy and bandwidth for each protocol. This reduces development
effort, and enables applications to use the “latest and greatest”
technologies on Linux or Microsoft Windows without the need
to re-compile and re-link applications.
Platform MPI is optimized for both distributed (DMP) and shared
memory (SMP) environments and provides a variety of flexible
CPU binding strategies for processes and threads, enabling bet-
ter performance on multi–core environments. With this capa-
bility memory and cache conflicts are managed by more intel-
ligently distributing the load among multiple cores.
With support for Windows HPC Server 2008 and the Microsoft job
scheduler, as well as other Microsoft operating environments,
Platform MPI allows developers targeting Windows platforms to
enjoy the benefits of a standard portable MPI and avoid propri-
etary lock-in.
WorkloadManagement

22 23
Features and Benefits
Features Benefits
Simplicity
ʎʎ Fully complies with the MPI 2.2 standard, providing dynamic processes,
one–sided communications, extended collectives, thread safety, and
updated ROMIO
ʎʎ Complete debugging, diagnostic and profiling tools
ʎʎ Auto-detection of interconnects and dynamic loading of libraries
ʎʎ No re-link required for debugging and profiling
ʎʎ Supported by the largest dedicated HPC support organization
ʎʎ Applications port easily to other platforms
ʎʎ Protects ISV software investment
ʎʎ Reduces time-to-market
ʎʎ Increased robustness and quality of applications
ʎʎ Technical problems resolved quickly and efficiently
Performance ʎʎ Improved shared memory performance, incorporating code and meth-
ods from Platform MPI 5.6 (Scali MPI)
ʎʎ 75% reduction in job startup and shutdown at scale
ʎʎ Scalability to 17,000 ranks
ʎʎ RDMA message progression & coalescing enhancements
ʎʎ Flexible CPU binding options maximize cache effectiveness and bal-
ance applications to minimize latency
ʎʎ Automated benchmarking of collective operations
ʎʎ Takes maximum advantage of available hardware
ʎʎ Reduced latency for better performance
ʎʎ Performance improves without explicit developer action
ʎʎ Better message throughput in streaming applications
ʎʎ Easier to optimize application performance
Compatibility ʎʎ Common source-code base between Linux and Windows
ʎʎ Binary compatible with applications developed for HP-MPI
ʎʎ MPICH-2 compatibility mode
ʎʎ Linux Standard Bindings ensure full compatibility across all major
Linux distributions
ʎʎ Scheduler agnostic with workload manager integrations for Windows
HPC, Platform LSF, PBS Pro, SLURM and other popular schedulers and
resource managers
ʎʎ Avoid the cost of separate releases for different platforms
ʎʎ Easily used with existing MPI applications
ʎʎ Common mpirun syntax between Linux and Windows
ʎʎ Customers avoid proprietary “lock-in”
ʎʎ Avoid floating point issues causing inconsistent results
Flexibility
ʎʎ Supports the widest variety of networks and interconnects
ʎʎ Select interconnects at run-time with no need to re-compile
ʎʎ Write applications once and deploy across multiple OS and hardware
topologies
ʎʎ CPU binding features well suited to GPU-aware applications
ʎʎ Develop applications that will run on more platforms
ʎʎ Reduce testing, maintenance and support costs
ʎʎ Enjoy strategic flexibility
WorkloadManagement
Supported Operating Systems
ʎʎ Red Hat Enterprise Linux 4.6, 5.x and 6.x
ʎʎ SUSE Linux Enterprise Server 10 and 11
ʎʎ CentOS 5.3
ʎʎ Microsoft Windows® XP/Vista, Server 2003/Server 200 8/HPC
Server 2008, Windows 7
Supported Interconnects and Protocols
Myrinet (Linux) GM & MX on X86-64 and Itanium2
InfiniBand (Linux) OFED, PSM, uDAPL on X86-64 and
Itanium2
OFED 1.1, 1.2, 1.3, 1.4, 1.5
SDR, DDR, QDR, ConnectX and
ConnextX2 Mellanox FCA
GigE (Linux) RDMA, uDAPL, TCP/IP
Infiniband (Windows) WinOF 2.x, IBAL, WSD, SDR, DDR,
QDR, ConnectX(2)
GigE (Windows) TCP/IP on x86-64

transtec HPC as a Service
You will get a range of applications like LS-Dyna, ANSYS,
Gromacs, NAMD etc. from all kinds of areas pre-installed,
integrated into an enterprise-ready cloud and workload
management system, and ready-to run. Do you miss your
application?
Ask us: HPC@transtec.de
transtec Platform as a Service
You will be provided with dynamically provided compute
nodes for running your individual code. The operating
system will be pre-installed according to your require-
ments. Common Linux distributions like RedHat, CentOS,
or SLES are the standard. Do you need another distribu-
tion?
Ask us: HPC@transtec.de
transtec Hosting as a Service
You will be provided with hosting space inside a profes-
sionally managed and secured datacenter where you
can have your machines hosted, managed, maintained,
according to your requirements. Thus, you can build up
your own private cloud. What range of hosting and main-
tenance services do you need?
Tell us: HPC@transtec.de
Services and Customer Care from A to Z
individual Presales
consulting
application-,
customer-,
site-speciﬁc
sizing of
HPC solution
burn-in tests
of systems
benchmarking of
different systems
continual
improvement
software
& OS
installation
application
installation
onsite
hardware
assembly
integration
into
customer’s
environment
customer
training
maintenance,
support &
managed services
individual Presales
consulting
application-,
customer-,
site-speciﬁc
sizing of
HPC solution
burn-in tests
of systems
benchmarking of
different systems
continual
improvement
software
& OS
installation
application
installation
onsite
hardware
assembly
integration
into
customer’s
environment
customer
training
maintenance,
support &
managed services
24 25
HPC @ transtec: Services and Customer Care from
A to Z
transtec AG has over 30 years of experience in scientific computing
and is one of the earliest manufacturers of HPC clusters. For nearly a
decade, transtec has delivered highly customized High Performance
clusters based on standard components to academic and industry
customers across Europe with all the high quality standards and the
customer-centricapproachthattransteciswellknownfor.
Every transtec HPC solution is more than just a rack full of hardware
–itisacomprehensivesolutionwitheverythingtheHPCuser,owner,
andoperatorneed.
In the early stages of any customer’s HPC project, transtec experts
provide extensive and detailed consulting to the customer – they
benefit from expertise and experience. Consulting is followed by
benchmarking of different systems with either specifically crafted
customer code or generally accepted benchmarking routines; this
aids customers in sizing and devising the optimal and detailed HPC
configuration.
EachandeverypieceofHPChardwarethatleavesourfactoryunder-
goes a burn-in procedure of 24 hours or more if necessary. We make
sure that any hardware shipped meets our and our customers’ qual-
ityrequirements.transtecHPCsolutionsareturnkeysolutions.Byde-
fault, a transtec HPC cluster has everything installed and configured
– from hardware and operating system to important middleware
components like cluster management or developer tools and the
customer’s production applications. Onsite delivery means onsite
integration into the customer’s production environment, be it estab-
lishing network connectivity to the corporate network, or setting up
softwareandconfigurationparts.
transtec HPC clusters are ready-to-run systems – we deliver, you turn
the key, the system delivers high performance. Every HPC project en-
tailstransfertoproduction:IToperationprocessesandpoliciesapply
to the new HPC system. Effectively, IT personnel is trained hands-on,
introduced to hardware components and software, with all opera-
tionalaspectsofconfigurationmanagement.
transtec services do not stop when the implementation projects
ends. Beyond transfer to production, transtec takes care. transtec of-
fers a variety of support and service options, tailored to the custom-
er’sneeds.Whenyouareinneedofanewinstallation,amajorrecon-
figuration or an update of your solution – transtec is able to support
your staff and, if you lack the resources for maintaining the cluster
yourself,maintaintheHPCsolutionforyou.FromProfessionalServic-
es to Managed Services for daily operations and required service lev-
els,transtecwillbeyourcompleteHPCserviceandsolutionprovider.
transtec’shighstandardsofperformance,reliabilityanddependabil-
ityassureyourproductivityandcompletesatisfaction.
HighPerformanceComputing
Performance Turns Into Productivity transtec’s offerings of HPC Managed Services offer customers the
possibility of having the complete management and administration
of the HPC cluster managed by transtec service specialists, in an ITIL
compliant way. Moreover, transtec’s HPC on Demand services help
provideaccesstoHPCresourceswhenevertheyneedthem,forexam-
ple, because they do not have the possibility of owning and running
an HPC cluster themselves, due to lacking infrastructure, know-how,
oradminstaff.
transtecHPCCloudServices
Last but not least transtec’s services portfolio evolves as customers‘
demands change. Starting this year, transtec is able to provide HPC
CloudServices.transtecusesadedicateddatacentertoprovidecom-
puting power to customers who are in need of more capacity than
they own, which is why this workflow model is sometimes called
computing-on-demand. With these dynamically provided resources,
customers with the possibility to have their jobs run on HPC nodes
in a dedicated datacenter, professionally managed and secured, and
individually customizable. Numerous standard applications like AN-
SYS,LS-Dyna,OpenFOAM,aswellaslotsofcodeslikeGromacs,NAMD,
VMD,andothersarepre-installed,integratedintoanenterprise-ready
cloudandworkloadmanagementenvironment,andreadytorun.
Alternatively, whenever customers are in need of space for hosting
theirownHPCequipmentbecausetheydonothavethespacecapac-
ity or cooling and power infrastructure themselves, transtec is also
able to provide Hosting Services to those customers who’d like to
have their equipment professionally hosted, maintained, and man-
aged.Customerscanthusbuilduptheirownprivatecloud!
Areyouinterestedinanyoftranstec’sbroadrangeofHPCrelatedser-
vices? Write us an email to HPC@transtec.de. We’ll be happy to hear
fromyou!

27
Scalable&Energy
EfficientHPCSystems
Data Centers today increas-
ingly face challenges to keep
pace with growth and vary-
ing business demands that
require rapid scaling and
technology evolution. While
each technology refresh cy-
cle delivers a significant per-
formance boost at roughly
the same price, these data
centers face rising facilities
costs and mounting energy
bills for power and cooling.
The operational costs for la-
bor and the ongoing equip-
ment maintenance costs are
other key components of
the Data Center Total Cost
of Ownership (TCO). In or-
der to get the most out of
their current and future IT
investments, Data Center
managers must minimize
TCO while achieving desired
efficiencies, computing per-
formance and effective man-
ageability of massive scale-
out environments.

28 29
Introduction
As a leading provider of High Performance Computing (HPC) solu-
tions across a wide range of application and industry segments
over the last few decades, IBM has constantly innovated and
helped alleviate customer pain points through its rich portfolio
for HPC. The new system design of IBM’s iDataPlex dx360 M4 serv-
ers leverages Intel’s industry-standard highly scalable, powerful
processors to address energy efficiency and significantly reduce
carbon emission footprints of next generation data centers. In-
tel’s Xeon processor E5-2600 product family using 32nm Sandy
Bridge processors power the IBM iDataPlex dx360 M4 air-cooled
half depth servers to provide more rack density per square feet
of floor space in the data center enabling higher compute power
for less energy per square foot of data center floor-space.
Overcoming Data Center Challenges
Business users, engineers and scientists attempting to solve
challenging problems in engineering, financial markets, me-
dia, public sector, oil & gas and the life and earth sciences, rely
heavily on HPC systems. These users are some of the most de-
manding IT solution clients. Their insatiable demand for com-
putational performance continues to drive HPC platforms to-
wards increasing system scalability, larger computing density,
lower power consumption, efficient system cooling, easier and
more reliable system management, and adaptable program-
ming environments.
Multiple industry studies indicate that revenue from HPC serv-
ers will continue to grow much faster than overall server rev-
enues with clusters being the dominant platform – over 75 per-
cent of HPC servers. The battle for HPC leadership will become
increasingly global and strategic. More real-world applications
will run at trans-petaflop speeds. However, these studies also
suggest that the escalating costs of server management, power
and cooling, reliability, availability and serviceability (RAS) of
HPC clusters with thousands of racks and facilities manage-
ment at HPC data centers will far outpace the costs of buying
new servers. This has caused a severe crisis in HPC data centers.
IT solution providers such as IBM have responded by designing
innovative solutions that address these issues while retaining
all the attractive attributes of industry-standard cluster archi-
tectures for HPC.
HPC Cluster Differentiators: iDataPlex Innovations in Packag-
ing, Power and Cooling
IBM’s iDataPlex dx360 M4 server can significantly reduce data
center costs arising from power consumption. It also helps re-
duce floor space requirements for cold aisles through innova-
tions in cooling technologies and smart system designs that
require less floor space. Data centers can deploy the next gen-
eration dx360 M4 with Intel Xeon processors E5-2600 product
family and obtain more computing density while consuming
the same or lower power. They also benefit from new pack-
aging and cooling methods – traditionally used in very large
enterprise class systems and proven very effective for design-
ing energy-efficient clusters where water is pumped through
the rack and into node-level cooling loops, which will provide
data centers with un-paralleled advantages.
The iDataPlex dx360 M4 has a hybrid air and direct water-
cooled design. The direct water-cooled solution is innovative
node level cooling that pumps water through the rack and into
node-level cooling loops. This water-cooled design eliminates
the need for fans in a 2U chassis and significantly boosts en-
ergy efficiency. This allows operations that are more reliable
and better placement of large clusters at HPC data centers –
significantly reducing the overall power, cooling and facilities
management costs for the entire data center.
iDataPlex: Scalable, Energy Efficient and Economical HPC
Workhorse
With direct water-cooling technology, the iDataPlex can help
lower power consumption by up to 40% and can greatly re-
duce the air conditioning need in data centers. Another ad-
vantage is reuse of 90% of the heat removed using direct
water-cooling technology from the data center for commu-
nity purposes such as heating buildings or water. The direct
effect of this innovative architecture is that it economically
increases the compute density in the data center by a factor
of five while retaining many of the attractive attributes of cur-
rent integrated rack cluster solutions from IBM. As compared
to air-cooling, direct water-cooling technology also consumes
much less energy.
IBM Continues Active Collaboration with HPC Developers and
Partners
Since the iDataPlex can provide a large scale-out cluster using
industry standard components, the large portfolio of HPC ap-
plications that have been mapped, migrated, and optimized
for the IBM Cluster architecture will benefit greatly from this
increased density, scalability, performance and cost-effective-
ness. This helps to protect customer’s prior investment in ap-
plication enablement and related. The iDataPlex provides an
efficient, reliable and cost-effective building block for super-
computing HPC clusters.
Scalable,EnergyEfficient
HPCSystems
IBM iDataPlex dx360 m4 HPC Clusters

IBM System x iDataPlex DIRECT WATER COOLED dx360 m4
Highlights:
ʎʎ 40% efficiency advantage over air cooled systems by elimi-
nating 90% of heat per node
ʎʎ Provides up to 10% power advantage over air-cooled sys-
tems
ʎʎ Two water cooling loops per node to cool processors and
memory dimms remaining components air cooled
ʎʎ 90% heat recovery can be reused for economical purposes,
such as warming other buildings and facilities
ʎʎ Typical operating conditions: Ambient air inlet @ 27-35°C
and water inlet @ 40-45°C to the nodes
30 31
iDataPlex: For High Performance and a Flexible
Data Center
To address the new data center challenges, a new way of de-
signing data centers and the server infrastructure that goes
into the data centers is essential. The design must encompass
data center power, cooling, management, and acquisition
as well as operating costs as the chief design goals, in addi-
tion to the performance, cost, and scalability requirements at
the server level. While IBM has years of experience designing
server technologies for scale-up and scale-out that primarily
focus on performance and scalability as the fundamental re-
quirements, the iDataPlex solution focuses on a different set
of goals:
ʎʎ Reduce the initial hardware acquisition costs and on-going
maintenance costs for data center owners
ʎʎ Improve efficiency in power consumption
ʎʎ Eliminate data center cooling requirements
ʎʎ Achieve higher server density within the same footprint as
the traditional rack layout
ʎʎ Simplify manageability for massive scale-out environments
ʎʎ Reduce the time to deployment through pre-configuration
and full integration at manufacturing
As is evident, these design goals go far beyond a single server or
a single rack level; they are goals for the entire data center. With
this new philosophy and the new design, the iDataPlex solution
promises to address the data center challenges at various levels:
ʎʎ an innovative rack design for higher node density
ʎʎ a flex node chassis based on industry standard components
ʎʎ shared power and cooling components to improve node as
well as rack level efficiencies
ʎʎ optional rear-door heat exchangers that virtually eliminates
traditional cooling and lastly,
ʎʎ optional direct water cooling technology that can lower pow-
er consumption by up to 40%
iDataPlex Architecture: Cost Effective, Flexible and Scalable
With major cooling innovations, the direct water-cooled
iDataPlex promises to be a unique offering for HPC data centers.
IBM’s enterprise experience and traditional HPC needs combine
to create a low cost platform with no redundant hardware, eas-
ily available industry standard parts, extreme compute density,
increased performance/watt, and quick and large-scale deploy-
ment. With a more efficient standard form factor, the iDataPlex
combines servers, chassis, switches, PDUs, management appli-
ances, and efficient rear door heat exchanger and latest direct
water-cooling technology to address many HPC needs such as
flexible configurations for nodes, storage and I/O capability in-
cluding power-optimized architectures.
With the Intel Xeon processor E5-2600 product family powered
iDataPlex M4, Petascale HPC clusters and data centers can
achieve over 55% reduction in overall TCO and as much as 90%
savings in power and cooling through innovations in air cooling,
rear door heat exchanger chassis, energy efficient Intel proces-
sors, rack, and optional liquid cooling in an iDataPlex.
iDataPlex Features and Benefits at a Glance
The iDataPlex is a result of a decade of system innovation by IBM
and promises to close the loop between low-end inexpensive
volume servers and enterprise class SMP and blade servers. This
massive, scalable computing workhorse includes a wide-range
of flexible options that make it a great choice for HPC applica-
HPCSystems

32 33
HPCSystems
tions and other large-scale applications. The figure below de-
picts the salient system details and summarizes the iDataPlex
features and benefits.
The iDataPlex: Tailored for HPC Applications
HPC users have always demanded computing solutions that
have the best performance, price/performance, and now are in-
creasingly demanding energy-efficient platforms. The iDataPlex
with its economical, large scale-out architecture is an excellent
workhorse for a wide-range of HPC applications in engineering,
life sciences, upstream petroleum, financial services, and scien-
tific research. Often, these applications do not require hardware
redundancy for failure protection, and software fault-tolerance
built into the application and/or system software is more than
adequate for practical purposes.
Over the last decade, in conjunction with application develop-
ers, IBM has made substantial investments to optimize and tune
HPC applications to take advantage of the IBM Cluster architec-
ture. Together they have achieved substantial performance tun-
ing by careful load balancing, maximizing single socket perfor-
mance, maintaining data locality, minimizing cache misses, and
maximizing the computation-to-communication ratio. In many
cases, application performance is further enhanced by the use of
optimized mathematical libraries and high-performance I/O so-
lutions like the IBM General Parallel File System (GPFS). All these
investments translate readily to the iDataPlex M4.
ʎʎ Increase data center density by eliminating hot/cold aisles
ʎʎ Eliminate rack heat exhaust
ʎʎ Same dimensions as standard iDataPlex rear door 4” deep
ʎʎ Liquid cooling at the rack is up to 95% more efficient than air
cooling by a CRAC
ʎʎ No electrical or moving parts
ʎʎ No condensation
ʎʎ Chilled water
ʎʎ Node-level cooling loops cool processors and memory, all
other components air cooled
ʎʎ Water manifolds for intra-rack distribution
ʎʎ 40% efficiency advantage over air cooled servers
ʎʎ 90% heat recover per node
ʎʎ Heat removed from the datacenter can be reused for
community purposes (ex. heating buildings or water)
The following figure highlights evolution of cooling technologies in IBM’s iDataPlex solution
Cooling at Rack Level with Rear
Door Heat Exchanger
Cooling at Node Level with Direct
Water Cooling&Today Tomorrow

34 35
HPCSystems
Benefits iDataPlex Features
Deployable
ʎʎ Racks fully integrated by IBM manufacturing at the factory site
ʎʎ Delivered, installed, and powered up at the Data Center before customer acceptance
ʎʎ Minimal on-site time required – shipping dock to image-load
Serviceable
ʎʎ All front access – eliminates need for accessing rear of rack
ʎʎ Blade-like design with chassis docking into power connector
ʎʎ Innovative cable layout provides highly efficient cable routing
ʎʎ Flexible support options from self maintenance to 24x7x4 response time
Efficient
ʎʎ Lower power consumption by up to 40% per rack
ʎʎ 90% extracted heat can be reused for community purposes (heating buildings)
ʎʎ Liquid cooling at the rack is up to 95% more efficient than air cooling by a CRAC
ʎʎ Energy expenses reduced by up to 40%
ʎʎ Minimize heat exhaust with optional Rear Heat Exchanger
Flexible
ʎʎ Excellent flexibility in node and rack configuration
ʎʎ Factory integrated racks delivered to the Data Center
ʎʎ 3rd party options and rack support
Affordable
ʎʎ Shared infrastructure designs cut cost out
ʎʎ Non-redundant components
ʎʎ TCO benefits -- Electricity cost ~90% lower, IT Cap Ex ~60% higher but 55% less overall TCO with iDataPlex.
Manageable
ʎʎ Blade like thinking in design
ʎʎ Scalable system management software for large-scale configurations
ʎʎ Rack management appliance for several racks
ʎʎ Ease in servicing from complete front access for upgrades
ʎʎ RAS advantages

36 37
Big Data, Cloud Storage it
doesn’t matter what you call
it, there is certainly increas-
ing demand to store larger
and larger amounts of un-
structured data.
The IBM General Parallel File
System (GPFSTM) has always
been considered a pioneer
of big data storage and con-
tinues today to lead in intro-
ducing industry leading stor-
age technologies. Since 1998
GPFS has lead the industry
with many technologies that
make the storage of large
quantities of file data pos-
sible. The latest version con-
tinues in that tradition, GPFS
3.5 represents a significant
milestone in the evolution of
big data management. GPFS
3.5 introduces some revolu-
tionary new features that
clearly demonstrate IBM’s
commitment to providing in-
dustry leading storage solu-
tions.
GeneralParallel
FileSystem(GPFS)

38 39
What is GPFS?
GPFS is more than clustered file system software; it is a full
featured set of file management tools. This includes advanced
storage virtualization, integrated high availability, automated
tiered storage management and the performance to effectively
manage very large quantities of file data.
GPFS allows a group of computers concurrent access to a com-
mon set of file data over a common SAN infrastructure, a net-
work or a mix of connection types. The computers can run any
mix of AIX, Linux or Windows Server operating systems. GPFS
provides storage management, information life cycle manage-
ment tools, centralized administration and allows for shared
access to file systems from remote GPFS clusters providing a
global namespace.
A GPFS cluster can be a single node, two nodes providing a high
availability platform supporting a database application, for
example, or thousands of nodes used for applications like the
modeling of weather patterns. The largest existing configura-
tions exceed 5,000 nodes. GPFS has been available on since 1998
and has been field proven for more than 14 years on some of
the world’s most powerful supercomputers2 to provide reliabil-
ity and efficient use of infrastructure bandwidth.
GPFS was designed from the beginning to support high per-
formance parallel workloads and has since been proven very
effective for a variety of applications. Today it is installed in
clusters supporting big data analytics, gene sequencing, digi-
tal media and scalable file serving. These applications are used
across many industries including financial, retail, digital media,
biotechnology, science and government. GPFS continues to
push technology limits by being deployed in very demanding
large environments. You may not need multiple petabytes of
data today, but you will, and when you get there you can rest
assured GPFS has already been tested in these enviroments.
This leadership is what makes GPFS a solid solution for any size
application.
Supported operating systems for GPFS Version 3.5 include AIX,
Red Hat, SUSE and Debian Linux distributions and Windows
Server 2008.
The file system
A GPFS file system is built from a collection of arrays that contain
the file system data and metadata. A file system can be built from
a single disk or contain thousands of disks storing petabytes of
data. Each file system can be accessible from all nodes within
the cluster. There is no practical limit on the size of a file system.
The architectural limit is 299 bytes. As an example, current GPFS
customers are using single file systems up to 5.4PB in size and
others have file systems containing billions of files.
Application interfaces
Applications access files through standard POSIX file system in-
terfaces. Since all nodes see all of the file data applications can
scale-out easily. Any node in the cluster can concurrently read
or update a common set of files. GPFS maintains the coheren-
cy and consistency of the file system using sophisticated byte
range locking, token (distributed lock) management and journal-
ing. This means that applications using standard POSIX locking
semantics do not need to be modified to run successfully on a
GPFS file system.
In addition to standard interfaces GPFS provides a unique set of
extended interfaces which can be used to provide advanced ap-
plication functionality. Using these extended interfaces an appli-
cation can determine the storage pool placement of a file, create
a file clone and manage quotas. These extended interfaces pro-
vide features in addition to the standard POSIX interface.
Performance and scalability
GPFS provides unparalleled performance for unstructured data.
GPFS achieves high performance I/O by:
Striping data across multiple disks attached to multiple nodes.
ʎʎ High performance metadata (inode) scans.
ʎʎ Supporting a wide range of file system block sizes to match I/O
requirements.
ʎʎ Utilizing advanced algorithms to improve read-ahead and wri-
te-behind IO operations.
ʎʎ Using block level locking based on a very sophisticated scalable
token management system to provide data consistency while
allowing multiple application nodes concurrent access to the
files.
When creating a GPFS file system you provide a list of raw devices
andtheyareassignedtoGPFSasNetworkSharedDisks(NSD).Once
a NSD is defined all of the nodes in the GPFS cluster can access the
disk, using local disk connection, or using t-he GPFS NSD network
protocol for shipping data over a TCP/IP or InfiniBand connection.
GPFS token (distributed lock) management coordinates access to
NSD’s ensuring the consistency of file system data and metadata
when different nodes access the same file. Token management
responsibility is dynamically allocated among designated nodes
in the cluster. GPFS can assign one or more nodes to act as token
managers for a single file system. This allows greater scalabil-
ity when you have a large number of files with high transaction
workloads. In the event of a node failure the token management
responsibility is moved to another node.
GeneralParallelFileSystem(GPFS)
Technologies That Enable the Management
of Big Data

40 41
All data stored in a GPFS file system is striped across all of the disks
within a storage pool, whether the pool contains 2 LUNS or 2,000
LUNS.Thiswidedatastripingallowsyoutogetthebestperformance
for the available storage. When disks are added to or removed from
astoragepoolexistingfiledatacanberedistributedacrossthenew
storage to improve performance. Data redistribution can be done
automatically or can be scheduled. When redistributing data you
can assign a single node to perform the task to control the impact
onaproductionworkloadorhaveallofthenodesintheclusterpar-
ticipate in data movement to complete the operation as quickly as
possible. Online storage configuration is a good example of an en-
terprise class storage management feature included in GPFS.
To achieve the highest possible data access performance GPFS
recognizes typical access patterns including sequential, reverse se-
quential and random optimizing I/O access for these patterns.
Along with distributed token management, GPFS provides scalable
metadata management by allowing all nodes of the cluster access-
ingthefilesystemtoperformfilemetadataoperations.Thisfeature
distinguishes GPFS from other cluster file systems which typically
have a centralized metadata server handling fixed regions of the
file namespace. A centralized metadata server can often become
a performance bottleneck for metadata intensive operations, lim-
iting scalability and possibly introducing a single point of failure.
GPFS solves this problem by enabling all nodes to manage meta-
data.
Administration
GPFS provides an administration model that is easy to use and is
consistentwithstandardfilesystemadministrationpracticeswhile
providingextensionsfortheclusteringaspectsofGPFS.Thesefunc-
tions support cluster management and other standard file system
administration functions such as user quotas, snapshots and ex-
tended access control lists.
GPFSadministrationtoolssimplifycluster-widetasks.AsingleGPFS
commandcanperformafilesystemfunctionacrosstheentireclus-
ter and most can be issued from any node in the cluster. Option-
ally you can designate a group of administration nodes that can be
used to perform all cluster administration tasks, or only authorize a
single login session to perform admin commands cluster-wide. This
allows for higher security by reducing the scope of node to node
administrative access.
Rolling upgrades allow you to upgrade individual nodes in the clus-
ter while the file system remains online. Rolling upgrades are sup-
ported between two major version levels of GPFS (and service lev-
els within those releases). For example you can mix GPFS 3.4 nodes
with GPFS 3.5 nodes while migrating between releases.
Quotasenabletheadministratortomanagefilesystemusagebyus-
ers and groups across the cluster. GPFS provides commands to gen-
erate quota reports by user, group and on a sub-tree of a file system
called a fileset. Quotas can be set on the number of files (inodes)
and the total size of the files. New in GPFS 3.5 you can now define
user and group per fileset quotas which allows for more options in
quota configuration. In addition to traditional quota management,
the GPFS policy engine can be used query the file system metadata
and generate customized space usage reports.
An SNMP interface allows monitoring by network management
applications. The SNMP agent provides information on the GPFS
cluster and generates traps when events occur in the cluster. For
example, an event is generated when a file system is mounted or if
a node fails. The SNMP agent runs on Linux and AIX. You can moni-
tor a heterogeneous cluster as long as the agent runs on a Linux or
AIX node.
You can customize the response to cluster events using GPFS
callbacks. A callback is an administrator defined script that is ex-
ecuted when an event occurs, for example, when a file system is
un-mounted for or a file system is low on free space. Callbacks can
be used to create custom responses to GPFS events and integrate
these notifications into various cluster monitoring tools.
GPFS provides support for the Data Management API (DMAPI) in-
terface which is IBM’s implementation of the X/Open data storage
management API. This DMAPI interface allows vendors of storage
management applications such as IBM Tivoli® Storage Manager
(TSM) and High Performance Storage System (HPSS) to provide Hi-
erarchical Storage Management (HSM) support for GPFS.
GPFS supports POSIX and NFS V4 access control lists (ACLs). NFS v4
ACLs can be used to serve files using NFSv4, but can also be used in
other deployments, for example, to provide ACL support to nodes
running Windows. To provide concurrent access from multiple op-
erating system types GPFS allows you to run mixed POSIX and NFS
v4 permissions in a single file system and map user and group IDs
between Windows and Linux/UNIX environments.
File systems may be exported to clients outside the cluster
through NFS. GPFS is often used as the base for a scalable NFS
file service infrastructure. The GPFS clustered NFS (cNFS) feature
provides data availability to NFS clients by providing NFS service
continuation if an NFS server fails. This allows a GPFS cluster to
provide scalable file service by providing simultaneous access to a
common set of data from multiple nodes. The clustered NFS tools
include monitoring of file services and IP address fail over. GPFS
cNFS supports NFSv3 only. You can export a GPFS file system using
NFSv4 but not with cNFS.
of Big Data

42 43
Data availability
GPFS is fault tolerant and can be configured for continued ac-
cess to data even if cluster nodes or storage systems fail. This is
accomplished though robust clustering features and support for
synchronous and asynchronous data replication.
GPFS software includes the infrastructure to handle data consis-
tency and availability. This means that GPFS does not rely on ex-
ternal applications for cluster operations like node failover. The
clustering support goes beyond who owns the data or who has
access to the disks. In a GPFS cluster all nodes see all of the data
and all cluster operations can be done by any node in the clus-
ter with a server license. All nodes are capable of performing all
tasks. What tasks a node can perform is determined by the type
of license and the cluster configuration.
As a part of the built-in availability tools GPFS continuously mon-
itors the health of the file system components. When failures are
detected appropriate recovery action is taken automatically. Ex-
tensive journaling and recovery capabilities are provided which
maintain metadata consistency when a node holding locks or
performing administrative services fails.
Snapshots can be used to protect the file system’s contents
against a user error by preserving a point in time version of the
file system or a sub-tree of a file system called a fileset. GPFS
implements a space efficient snapshot mechanism that gener-
ates a map of the file system or fileset at the time the snaphot
is taken. New data blocks are consumed only when the file sys-
tem data has been deleted or modified after the snapshot was
created. This is done using a redirect-on-write technique (some-
times called copy-on-write). Snapshot data is placed in existing
storage pools simplifying administration and optimizing the use
of existing storage. The snapshot function can be used with a
backup program, for example, to run while the file system is in
use and still obtain a consistent copy of the file system as it was
when the snapshot was created. In addition, snapshots provide
an online backup capability that allows files to be recovered eas-
ily from common problems such as accidental file deletion.
Data Replication
For an additional level of data availability and protection synchro-
nous data replication is available for file system metadata and
data. GPFS provides a very flexible replication model that allows
you to replicate a file, set of files, or an entire file system. The rep-
lication status of a file can be changed using a command or by us-
ing the policy based management tools. Synchronous replication
allowsforcontinuousoperationevenifapathtoanarray,anarray
itself or an entire site fails.
Synchronous replication is location aware which allows you to
optimize data access when the replicas are separated across a
WAN. GPFS has knowledge of what copy of the data is “local” so
read-heavy applications can get local data read performance even
when data replicated over a WAN. Synchronous replication works
well for many workloads by replicating data across storage arrays
within a data center, within a campus or across geographical dis-
tances using high quality wide area network connections.
When wide area network connections are not high performance
or are not reliable, an asynchronous approach to data replication
is required. GPFS 3.5 introduces a feature called Active File Man-
agement (AFM). AFM is a distributed disk caching technology de-
veloped at IBM Research that allows the expansion of the GPFS
global namespace across geographical distances. It can be used to
provide high availability between sites or to provide local “copies”
of data distributed to one or more GPFS clusters. For more details
on AFM see the section entitled Sharing data between clusters.
For a higher level of cluster reliability GPFS includes advanced
clustering features to maintain network connections. If a network
connection to a node fails GPFS automatically tries to reestablish
the connection before marking the node unavailable. This can
provide for better uptime in environments communicating across
a WAN or experiencing network issues.
Using these features along with a high availability infrastructure
ensures a reliable enterprise class storage solution.
GPFS Native Raid (GNR)
Larger disk drives and larger file systems are creating challenges
for traditional storage controllers. Current RAID 5 and RAID 6
based arrays do not address the challenges of Exabyte scale stor-
age performance, reliability and management. To address these
challenges GPFS Native RAID (GNR) brings storage device man-
agement into GPFS. With GNR GPFS can directly manage thou-
sands of storage devices. These storage devices can be individual
disk drives or any other block device eliminating the need for a
storage controller.
GNR employs a de-clustered approach to RAID. The de-clustered
architecture reduces the impact of drive failures by spreading
data over all of the available storage devices improving appli-
cation IO and recovery performance. GNR provides very high
reliability through an 8+3 Reed Solomon based raid code that
divides each block of a file into 8 parts and associated parity.
This algorithm scales easily starting with as few as 11 storage
devices and growing to over 500 per storage pod. Spreading the
data over many devices helps provide predicable storage perfor-
mance and fast recovery times measured in minutes rather than
hours in the case of a device failure. In addition to performance
of Big Data

44 45
improvements GNR provides advanced checksum protection to
ensure data integrity. Checksum information is stored on disk
and verified all the way to the NSD client.
Information lifecycle management (ILM) toolset
GPFS can help you to achieve data lifecycle management efficien-
cies through policy-driven automation and tiered storage man-
agement. The use of storage pools, filesets and user-defined poli-
cies provide the ability to better match the cost of your storage to
the value of your data.
Storage pools are used to manage groups of disks within a file sys-
tem. Using storage pools you can create tiers of storage by group-
ing disks based on performance, locality or reliability characteris-
tics. For example, one pool could contain high performance solid
statedisk(SSD)disksandanothermoreeconomical7,200RPMdisk
storage. These types of storage pools are called internal storage
pools. When data is placed in or moved between internal storage
pools all of the data management is done by GPFS. In addition to
internal storage pools GPFS supports external storage pools. Ex-
ternal storage pools are used to interact with an external storage
management application including IBM Tivoli® Storage Manager
(TSM) and High Performance Storage System (HPSS). When moving
data to an external pool GPFS handles all of the metadata process-
ing then hands the data to the external application for storage on
alternate media, tape for example. When using TSM or HPSS data
can be retrieved from the external storage pool on demand, as a
result of an application opening a file or data can be retrieved in a
batch operation using a command or GPFS policy.
A fileset is a sub-tree of the file system namespace and provides
a way to partition the namespace into smaller, more manageable
units. Filesets provide an administrative boundary that can be
used to set quotas, take snapshots, define AFM relationships and
be used in user defined policies to control initial data placement
or data migration. Data within a single fileset can reside in one or
more storage pools. Where the file data resides and how it is man-
aged once it is created is based on a set of rules in a user defined
policy.
There are two types of user defined policies in GPFS: file place-
ment and file management. File placement policies determine in
whichstoragepoolfiledataisinitiallyplaced.Fileplacementrules
are defined using attributes of a file known when a file is created
such as file name, fileset or the user who is creating the file. For
example a placement policy may be defined that states ‘place all
files with names that end in .mov onto the near-line SAS based
storage pool and place all files created by the CEO onto the SSD
based storage pool’ or ‘place all files in the fileset ‘development’
onto the SAS based storage pool’.
Once files exist in a file system, file management policies can be
used for file migration, deletion, changing file replication status
or generating reports.
You can use a migration policy to transparently move data from
one storage pool to another without changing the file’s location
in the directory structure. Similarly you can use a policy to change
the replication status of a file or set of files, allowing fine grained
control over the space used for data availability.
You can use migration and replication policies together, for exam-
ple a policy that says: ‘migrate all of the files located in the subdi-
rectory /database/payroll which end in *.dat and are greater than
1 MB in size to storage pool #2 and un-replicate these files’.
File deletion policies allow you to prune the file system, deleting
files as defined by policy rules. Reporting on the contents of a file
system can be done through list policies. List policies allow you to
quickly scan the file system metadata and produce information
listing selected attributes of candidate files.
Filemanagementpoliciescanbebasedonmoreattributesofafile
than placement policies because once a file exists there is more
known about the file. For example file placement attributes can
utilize attributes such as last access time, size of the file or a mix
of user and file size. This may result in policies like: ‘Delete all files
with a name ending in .temp that have not been accessed in the
last 30 days’, or ‘Migrate all files owned by Sally that are larger
than 4GB to the SATA storage pool’.
Rule processing can be further automated by including attri-
butes related to a storage pool instead of a file using the thresh-
old option. Using thresholds you can create a rule that moves
files out of the high performance pool if it is more than 80%
full, for example. The threshold option comes with the ability to
set high, low and pre-migrate thresholds. Pre-migrated files are
files that exist on disk and are migrated to tape. This method
is typically used to allow disk access to the data while allow-
ing disk space to be freed up quickly when a maximum space
threshold is reached. This means that GPFS begins migrating
data at the high threshold, until the low threshold is reached.
If a pre-migrate threshold is set GPFS begins copying data un-
til the pre-migrate threshold is reached. This allows the data to
continue to be accessed in the original pool until it is quickly
deleted to free up space the next time the high threshold is
reached. Thresholds allow you to fully utilize your highest per-
formance storage and automate the task of making room for
new high priority content.
Policy rule syntax is based on the SQL 92 syntax standard and
supports multiple complex statements in a single rule enabling
powerful policies. Multiple levels of rules can be applied to a
of Big Data

Figure 1: SAN attached Storage
46 47
file system, and rules are evaluated in order for each file when
the policy engine executes allowing a high level of flexibility.
GPFS provides unique functionality through standard interfaces,
an example of this is extended attributes. Extended attributes
are a standard POSIX facility.
GPFS has long supported the use of extended attributes, though
in the past they were not commonly used, in part because of per-
formance concerns. In GPFS 3.4, a comprehensive redesign of the
extended attributes support infrastructure was implemented,
resulting in significant performance improvements. In GPFS 3.5,
extended attributes are accessible by the GPFS policy engine al-
lowing you to write rules that utilize your custom file attributes.
Executing file management operations requires the ability to
efficiently process the file metadata. GPFS includes a high per-
formance metadata scan interface that allows you to efficiently
process the metadata for billions of files. This makes the GPFS
ILM toolset a very scalable tool for automating file management.
This high performance metadata scan engine employs a scale-
out approach. The identification of candidate files and data
movement operations can be performed concurrently by one or
more nodes in the cluster. GPFS can spread rule evaluation and
data movement responsibilities over multiple nodes in the clus-
ter providing a very scalable, high performance rule processing
engine.
Cluster configurations
GPFS supports a variety of cluster configurations independent
of which file system features you use. Cluster configuration op-
tions can be characterized into three basic categories:
ʎʎ Shared disk
ʎʎ Network block I/O
ʎʎ Synchronously sharing data between clusters.
ʎʎ Asynchronously sharing data between clusters.
Shared disk
A shared disk cluster is the most basic environment. In this con-
figuration, the storage is directly attached to all machines in the
cluster as shown in Figure 1. The direct connection means that
each shared block device is available concurrently to all of the
nodes in the GPFS cluster. Direct access means that the storage
is accessible using a SCSI or other block level protocol using a
SAN, Infiniband, iSCSI, Virtual IO interface or other block level IO
connection technology.
Figure 1 illustrates a GPFS cluster where all nodes are connect-
ed to a common fibre channel SAN. This example shows a fibre
channel SAN though the storage attachment technology could
be InfiniBand, SAS, FCoE or any other.. The nodes are connected
to the storage using the SAN and to each other using a LAN. Data
used by applications running on the GPFS nodes flows over the
SAN and GPFS control information flows among the GPFS in-
stances in the cluster over the LAN.
This configuration is optimal when all nodes in the cluster need
the highest performance access to the data. For example, this is
a good configuration for providing etwork file service to client
systems using clustered NFS, high-speed data access for digital
media applications or a grid infrastructure for data analytics.
Network-based block IO
As data storage requirements increase and new storage and con-
nection technologies are released a single SAN may not be a suf-
ficient or appropriate choice of storage connection technology. In
environments where every node in the cluster is not attached to a
single SAN, GPFS makes use of an integrated network block device
capability. GPFS provides a block level interface over TCP/IP net-
works called the Network Shared Disk (NSD) protocol. Whether us-
ing the NSD protocol or a direct attachment to the SAN the mount-
edfilesystemlooksthesametotheapplication,GPFStransparently
handles I/O requests.
GPFS clusters can use the NSD protocol to provide high speed
data access to applications running on LAN-attached nodes. Data
is served to these client nodes from one or more NSD servers. In
this configuration, disks are attached only to the NSD servers. Each
NSD server is attached to all or a portion of the disk collection. With
GPFSyoucandefineuptoeightNSDserversperdiskanditisrecom-
mended that at least two NSD servers are defined for each disk to
avoid a single point of failure.
GPFS uses the NSD protocol over any TCP/IP capable network fab-
ric.OnLinuxGPFScanusetheVERBSRDMAprotocoloncompatible
fabrics (such as InfiniBand) to transfer data to NSD clients. The net-
of Big Data

Figure 2: Network block IO
48 49
The choice of how many nodes to configure as NSD servers is based
onperformancerequirements,thenetworkarchitectureandtheca-
pabilities of the storage subsystems. High bandwidth LAN connec-
tions should be used for clusters requiring significant data transfer
rates. This can include 1Gbit or 10 Gbit Ethernet. For additional per-
formance or reliability you can use link aggregation (EtherChannel
or bonding), networking technologies like source based routing or
higher performance networks such as InfiniBand.
ThechoicebetweenSANattachmentandnetworkblockI/Oisaper-
formance and economic one. In general, using a SAN provides the
work fabric does not need to be dedicated to GPFS; but should pro-
videsufficientbandwidthtomeetyourGPFSperformanceexpecta-
tionsandforapplicationswhichsharethebandwidth.GPFShasthe
ability to define a preferred network subnet topology, for example
designate separate IP subnets for intra-cluster communication and
the public network. This provides for a clearly defined separation of
communication traffic and allows you to increase the throughput
andpossiblythenumberofnodesinaGPFScluster.Allowingaccess
to the same disk from multiple subnets means that all of the NSD
clients do not have to be on a single physical network. For example
you can place groups of clients onto separate subnets that access
a common set of disks through different NSD servers so not all NSD
servers need to serve all clients. This can reduce the networking
hardware costs and simplify the topology reducing support costs,
providing greater scalability and greater overall performance.
An example of the NSD server model is shown in Figure 2. In this
configuration, a subset of the total node population is defined as
NSDserver nodes. The NSD Server is responsible for the abstraction
of disk datablocks across a TCP/IP or Infiniband VERBS (Linux only)
based network. The factthat the disks are remote is transparent
to the application. Figure 2 shows an example of a configuration
where a set of compute nodes are connected to a set of NSD serv-
ers using a high-speed interconnect or an IP-based network such
as Ethernet. In this example, data to the NSD servers flows over
the SAN and both data and control information to the clients flow
across the LAN. Since the NSD servers are serving data blocks from
one or more devices data access is similar to a SAN attached envi-
ronment in that data flows from all servers simultaneously to each
client. This parallel data access provides the best possible through-
put to all clients. In addition it provides the ability to scale up the
throughput even to a common data set or even a single file.
highest performance; but the cost and management complexity of
SANs for large clusters is often prohibitive. In these cases network
block I/O provides an option.
Network block I/O is well suited to grid computing and clusters
with sufficient network bandwidth between the NSD servers and
the clients. For example, an NSD protocol based grid is effective for
web applications, supply chain management or modeling weather
patterns.
Mixed Clusters
The last two sections discussed shared disk and network attached
GPFS cluster topologies. You can mix these storage attachment
methods within a GPFS cluster to better matching the IO require-
ments to the connection technology.
A GPFS node always tries to find the most efficient path to the stor-
age. If a node detects a block device path to the data it is used. If
there is no block device path then the network is used. This capa-
bility can be leveraged to provide additional availability. If a node
is SAN attached to the storage and there is an HBA failure, for ex-
ample, GPFS can fail over to using the network path to the disk. A
mixed cluster topology can provide direct storage access to non-
NSD server nodes for high performance operations including back-
ups or data ingest.
Sharing data between clusters
TherearetwomethodsavailabletosharedataacrossGPFSclusters:
GPFS multi-cluster and a new feature called Active File Manage-
ment (AFM).
GPFS Multi-cluster allows you to utilize the native GPFS protocol to
share data across clusters. Using this feature you can allow other
clusters to access one or more of your file systems and you can
of Big Data

Figure 3: Mixed Cluster Atchitecture
Figure 4: Multi-cluster
50 51
could include a SAN. Figure 3 illustrates a multi-cluster configura-
tion with both LAN and mixed LAN and SAN connections.
In Figure 3, Cluster B and Cluster C need to access the data from
Cluster A. Cluster A owns the storage and manages the file system.
It may grant access to file systems which it manages to remote
clusters such as Cluster B and Cluster C. In this example, Cluster B
andClusterCdonothaveanystoragebutthatisnotarequirement.
They could own file systems which may be accessible outside their
cluster. Commonly in the case where a cluster does not own stor-
age, the nodes are grouped into clusters for ease of management.
When the remote clusters need access to the data, they mount the
file system by contacting the owning cluster and passing required
security checks. In Figure 3, Cluster B acesses the data through the
NSD protocol. Cluster C accesses data through an extension of the
SAN. For cluster C access to the data is similar to the nodes in the
host cluster. Control and administrative traffic flows through the IP
network shown in Figure 3 and data access is direct over the SAN.
Both types of configurations are possible and as in Figure 3 can be
mixed as required.
Multi-cluster environments are well suited to sharing data across
clusters belonging to different organizations for collaborative
computing , grouping sets of clients for administrative purposes or
implementing a global namespace across separate locations.
A multi-cluster configuration allows you to connect GPFS clusters
within a data center, across campus or across reliable WAN links.
For sharing data between GPFS clusters across less reliable WAN
links or in cases where you want a copy of the data in multiple loca-
tionsyoucanuseanewfeatureintroducedinGPFS3.5calledActive
File Management.
Active File Management (AFM) allows you to create associations
between GPFS clusters. Now the location and flow of file data be-
mount file systems that belong to other GPFS clusters for which
you have been authorized. A multi-cluster environment allows the
administratortopermitaccesstospecificfilesystemsfromanother
GPFS cluster. This feature is intended to allow clusters toshare data
athigherperformancelevelsthanfilesharingtechnologieslikeNFS
or CIFS. It is not intended to replace such file sharing technologies
which are optimized for desktop access or for access across unreli-
able network links.
Multi-cluster capability is useful for sharing across multiple clus-
terswithinaphysicallocationoracrosslocations.Clustersaremost
often attached using a LAN, but in addition the cluster connection
tween GPFS clusters can be automated. Relationships between
GPFS clusters using AFM are defined at the fileset level. A fileset in
a file system can be created as a “cache” that provides a view to a
file system in another GPFS cluster called the “home.” File data is
moved into a cache fileset on demand. When a file is read the in the
cache fileset the file data is copied from the home into the cache fi-
leset.Dataconsistencyandfilemovementintoandoutofthecache
of Big Data

Store 3
Store 1
Store 2
Cache:
/data1
Local:
/data1
Cache:
/data3
Cache:
/data5Cache:
/data1
Local:
/data3
Cache:
/data3
Cache:
/data5
Local:
/data5
Client access:
/global/data1
/global/data2
/global/data3
/global/data3
/global/data4
/global/data5
/global/data6
Client access:
/global/data1
/global/data2
/global/data3
/global/data3
/global/data4
/global/data5
/global/data6
Client access:
/global/data1
/global/data2
/global/data3
/global/data3
/global/data4
/global/data5
/global/data6
Figure 5: Global Namespace using AFM
52 53
is managed automatically by GPFS.
Cache filesets can be read-only or writeable. Cached data is local-
ly read or written. On read if the data is not in in the “cache” then
GPFS automatically creates a copy of the data. When data is writ-
ten into the cache the write operation completes locally then GPFS
asynchronously pushes the changes back to the home location.
You can define multiple cache filesets for each home data source.
The number of cache relationships for each home is limited only
by the bandwidth available at the home location. Placing a quota
on a cache fileset causes the data to be cleaned (evicted) out of the
cache automatically based on the space available. If you do not set
a quota a copy of the file data remains in the cache until manually
evicted or deleted.
YoucanuseAFMtocreateaglobalnamespacewithinadatacenter,
acrossacampusorbetweendatacenterslocatedaroundtheworld.
AFM is designed to enable efficient data transfers over wide area
network (WAN) connections. When file data is read from home into
cache that transfer can happen in parallel within a node called a
gateway or across multiple gateway nodes.
UsingAFMyoucannowcreateatrulyworld-wideglobalnamespace.
Figure 5 is an example of a global namespace built using AFM. In
this example each site owns 1/3 of the namespace, store3 “owns”
/data5 and /data6 for example. The other sites have cache filesets
that point to the home location for the data owned by the other
clusters. In this example store3 /data1-4 cache data owned by the
other two clusters. This provides the same namespace within each
GPFS cluster providing applications one path to a file across the en-
tire organization.
Youcanachievethistypeofglobalnamespacewithmulti-clusteror
AFM, which method you chose depends on the quality of the WAN
link and the application requirements.
of Big Data

54 55
in the past they were not commonly used, in part because of per-
formance concerns. In GPFS 3.4, a comprehensive redesign of the
extended attributes support infrastructure was implemented,
resulting in significant performance improvements. In GPFS 3.5,
extended attributes are accessible by the GPFS policy engine al-
lowing you to write rules that utilize your custom file attributes.
Now an application can use standard POSIX interfaces t to man-
age extended attributes and the GPFS policy engine can utilize
these attributes.
Independent Filesets
To effectively manage a file system with billions of files requires
advanced file management technologies. GPFS 3.5 introduces a
new concept called the independent fileset. An independent fi-
leset has its own inode space. This means that an independent
fileset can be managed similar to a separate file system but still
allow you to realize the benefits of storage consolidation.
An example of an efficiency introduced with independent file-
sets is improved policy execution performance. If you use an in-
dependent fileset as a predicate in your policy, GPFS only needs
to scan the inode space represented by that fileset, so if you
have 1 billion files in your file system and a fileset has an inode
space of 1 million files, the scan only has to look at 1 million in-
odes. This instantly makes the policy scan much more efficient.
Independent filesets enable other new fileset features in GPFS
3.5.
Fileset Level Snapshots
Snapshot granularity is now at the fileset level in addition to file
system level snapshots.
What’s new in GPFS Version 3.5
For those who are familiar with GPFS 3.4 this section provides a
list of what is new in GPFS Version 3.5.
Active File Management
When GPFS was introduced in 1998 it represented a revolution in
file storage. For the first time a group of servers could share high
performance access to a common set of data over a SAN or net-
work. The ability to share high performance access to file data
across nodes was the introduction of the global namespace.
Later GPFS introduced the ability to share data across multiple
GPFS clusters. This multi-cluster capability enabled data sharing
between clusters allowing for better access to file data. This fur-
ther expanded the reach of the global namespace from within a
cluster to across clusters spanning a data center or a country.
There were still challenges to building a multi-cluster global
namespace. The big challenge is working with unreliable and
high latency network connections between the servers. Active
File Management (AFM) in GPFS addresses the WAN bandwidth is-
sues and enables GPFS to create a world-wide global namespace.
AFM ties the global namespace together asynchronously provid-
ing local read and write performance with automated namespace
management. It allows you to create associations between GPFS
clusters and define the location and flow of file data.
High Performance Extended Attributes
GPFS has long supported the use of extended attributes, though
Fileset Level Quotas
User and group quotas can be set per fileset
File Cloning
File clones are space efficient copies of a file where two instanc-
es of a file share data they have in common and only changed
blocks require additional storage. File cloning is an efficient way
to create a copy of a file, without the overhead of copying all of
the data blocks.
IPv6 Support
IPv6 support in GPFS means that nodes can be defined using
multiple addresses, both IPv4 and IPv6.
GPFS Native RAID
GPFS Native RAID (GNR) brings storage RAID management into
the GPFS NSD server. With GNR GPFS directly manages JBOD
based storage. This feature provides greater availability, flexibil-
ity and performance for a variety of application workloads.
GNR implements a Reed-Solomon based de-clustered RAID tech-
nology that can provide high availability and keep drive failures
from impacting performance by spreading the recovery tasks
over all of the disks. Unlike standard Network Shared Disk (NSD)
data access GNR is tightly integrated with the storage hardware.
For GPFS 3.5 GNR is available on the IBM Power 775 Supercom-
puter platform.
What’s new in GPFS Version 3.5

58 59
Always keep in touch with the latest news
Visit us on the Web!
Here you will find comprehensive information about HPC, IT solutions for the datacenter, services and
high-performance, efficient IT systems.
Subscribe to our technology journals, E-News or the transtec newsletter and always stay up to date.
www.transtec.de/en/hpc
Simulation
Simulation
CAD
CAD
CAE
CAE
Engineering
Engineering
Price Modelling
ice Modelling
Life Sciences
Life Sciences
Automotive
Automotive
Aerospace
Aerospace
Risk Analysis
Risk Analysis
Big Data Analytics
Data Analytics

Simulation
CAD
CAEEngineering
Price Modelling
Life Sciences
Automotive
Aerospace
Risk Analysis
Risk Analysis
Big Data Analytics
60
transtec Germany
Tel +49 (0) 7071/703-400
transtec@transtec.de
www.transtec.de
transtec Switzerland
Tel +41 (0) 44/818 47 00
transtec.ch@transtec.ch
www.transtec.ch
transtec United Kingdom
Tel +44 (0) 1295/756 500
transtec.uk@transtec.co.uk
www.transtec.co.uk
ttec Netherlands
Tel +31 (0) 24 34 34 210
ttec@ttec.nl
www.ttec.nl
transtec France
Tel +33 (0) 3.88.55.16.00
transtec.fr@transtec.fr
www.transtec.fr
Texts and concept:
Layout and design:
Dr. Oliver Tennert, Director Technology Management & HPC Solutions | Oliver.Tennert@transtec.de
Jennifer Kemmler, Graphics & Design | Jennifer.Kemmler@transtec.de
© transtec AG, May 2013
The graphics, diagrams and tables found herein are the intellectual property of transtec AG and may be reproduced or published only with its express permission.
No responsibility will be assumed for inaccuracies or omissions. Other names or logos may be trademarks of their respective owners.

HPC Compass IBM Special 2013/14

More Related Content

What's hot (10)

Viewers also liked (7)

Similar to HPC Compass IBM Special 2013/14 (20)

Recently uploaded (20)

HPC Compass IBM Special 2013/14