Modular Computing

MODULAR COMPUTING

A SEMINAR REPORT

Submitted by
SREEKANTH K T

in partial fulfillment for the award of the degree

of

BTECH DEGREE

in

COMPUTER SCIENCE & ENGINEERING

SCHOOL OF ENGINEERING

COCHIN UNIVERSITY OF SCIENCE &
TECHNOLOGY KOCHI - 682022

SEPTEMBER 2010

Division of Computer Engineering
School of Engineering
Cochin University of Science & Technology
Kochi-682022
_________________________________________________________

CERTIFICATE

Certified that this is a bonafied record of the seminar work titled

Modular Computing

Done by

Sreekanth K T

of VI semester Computer Science & Engineering in the year 2010 in partial

fulfillment of the requirements for the award of Degree of Bachelor of Technology

in Computer Science & Engineering of Cochin University of Science & Technology

Dr. David Peter S Anupama V
Head of the Division Seminar Guide

ACKNOWLEDGEMENT

I thank GOD almighty for guiding me throughout the seminar. I would
like to thank all those who have contributed to the completion of the
seminar and helped me with valuable suggestions for improvement. I am
extremely grateful to Dr. David Peter, HOD, Division of Computer
Science, for providing me with best facilities and atmosphere for the
creative work guidance and encouragement. I would like to thank my
coordinator, Sudheep Elayidom, Sr. Lecturer, Division of Computer
Science, and my guide Anupama V , Lecturer , Division of Computer
Science, SOE for all help and support extend to me. I thank all Staff
members of my college and friends for extending their cooperation during
my seminar. Above all I would like to thank my parents without whose
blessings; I would not have been able to accomplish my goal.

SREEKANTH K T

ABSTRACT

The modular, brick-style technology stands to revolutionize the
way people buy high-performance computers, allowing them to expand
and upgrade only the elements they need for their systems or add new
technologies as they become available. Traditionally, users have had to
buy expensive "one size fits all" systems that either were too much for
their needs or became obsolete quickly and had to be replaced--a costly
and cumbersome process.
Without this modular approach, conventional high-performance
systems often need to be replaced as often as once a year to keep up
with changing needs, new technology or competitive pressures--at a
cost potentially in the millions of dollars for each replacement. This
daunting prospect can limit the progress of research and development
and can hold industries and scientific pursuits back.
Now, technical and creative computer users can have the same
modularity, freedom of choice, and ease of upgrade that people have
long benefited from in assembling and enhancing their home-
entertainment centers.

TABLE OF CONTENTS

CHAPTER NO. TITLE PAGE NO.

1. INTRODUCTION 1

2. PROBLEMS IN IT SCENARIO 2

2.1. COST OF OVER PROVISIONING 2

2.2. COST OF HIGH AVAILABILITY 3

2.3. COST OF TOO MANY PEOPLE DOING LOW LEVEL TASK 4

2.4. EXCESSIVE SERVER MANAGEMENT 4

2.5. EXCESSIVE DEPLOYMENT EXPENSES 4

2.6. EXCESSIVE CABLE MANAGEMENT 5

2.7. COST OF STRANDED RESOURCES 5

3. MODULAR COMPUTING ARCHITECTURE 6

3.1. PROCESSING RESOURCE POOL 7

3.2. STORAGE RESOURCE POOL 7

3.3. NETWORKING RESOURCE POOL 7

3.4. MODULAR COMPUTING SOFTWARE (MCS) 8

3.5. VIRTUAL SERVERS 8

4. ENTERPRISE MODULAR COMPUTING 9

4.1. A NEW COMPUTING PARADIGM 9

4.2. PROCESSING RESOURCE 9

4.3. STORAGE RESOURCE 10

4.4. NETWORKING RESOURCE 10

4.5. MODULAR COMPUTING SOFTWARE 11

4.6. FEWER CABLES TO MANAGE 11

5. BENEFITS OF MODULAR COMPUTING 12

5.1. WINNING THROUGH MODULARITY 12

5.2. DOING MORE WITH LESS 12

5.3. INCREASED AGILITY 12

5.4. REDUCED PEOPLE COST 13

5.5. REDUCED EQUIPMENT COST 14

6. APPLICATIONS 16

6.1. THE EGENERA BLADEFRAME SYSTEM 16

6.2. NUMA FLEX BY SGI 18

6.2.1. ABOUT THE TECHNOLOGY 19

6.2.2. EXECUTING ON SGI’S CORPORATE STRATEGY 20

6.2.3. A SUPERIOR ARCHITECTURE 21

6.2.4. WINNING THROUGH MODULARITY 22

6.2.5. THE POWER OF VISUALIZATION 23

7. REMARKS FROM EXPERTS 24

8. CONCLUSION 26

REFERENCES 27

MODULAR COMPUTING

1 INTRODUCTION

I T’s Challenge

In the past three years, the world has changed for information
technology groups. In the late 1990s, the predominant problem was
deploying equipment and software quickly enough to keep up with demand
for computing. While the tech sector boomed on Wall Street, money was no
object. IT budgets swelled and the numbers of computers in data centers
grew exponentially.
Now, in the early 2000s, the picture is very different. IT budgets are
flat down, yet business demand for IT services continues to escalate. This
combination of more demand and constrained budgets has compelled IT
groups to consider new approaches to IT infrastructure, approaches that offer
more flexibility and lower cost of ownership.
The common theme is cost cutting. In today’s world, profits come
less easily than in 1990s. Competitors are more experienced, and
competition is more intense. Corporations that trim costs while providing
great service will prevail over those that can’t.
IT plays a major role in this competitive situation. As competition
becomes more intense, so does the pressure on IT to cut costs and boost
contribution. Now more than ever, large corporations are using their
computing assets as tools to pull ahead of the competition.
The January 13, 2003, issue of Time Magazine provides a great
example of how IT contributes in new ways. Executives at a big-box retailer
were considering dropping a particular brand of chicken from the shelves
because the sales volume was poor. Then the retailer’s data miners found
that customers who bought that brand of chicken also bought large amounts
of other merchandise. The chicken stayed.

DIVISION OF COMPUTER SCIENCE ENGINEERING, SOE CUSAT 1

MODULAR COMPUTING

Data mining, online transactions and other new computing demands
require collecting and processing enormous amounts of data. Still, IT
departments are expected to keep up, even with budgets flat down. The
bottom line is that IT will be doing more with less.
Modular Computing can slash costs in IT infrastructure. It enables IT
groups to consolidate equipment, conserving expensive real estate. It offers
the opportunity to migrate applications from expensive proprietary platforms
to more, powerful, and manageable systems.

2 PROBLEMS IN IT SCENARIO

2.1 Cost of Over-Provisioning
As data centers have moved toward servers and away mainframes, IT
has found that some mainframe capabilities weren’t available on servers. A
glaring example is that smaller servers were unable to rapidly obtain more
processing power to accommodate peaks in computing demand.
As applications became more transactional, for example with
customers entering information via the Web, these peaks in computing
demand became more visible. During peak demand, customers saw their
transactions slow down. In situations where these transactions affect the
bottom line, as when customers enter purchases, prompt processing becomes
vital to the business.
As the number of customers using Web services has increased, the
peaks in computing demand became more intense and more frequent.
Consequently, customers more frequently saw declines in performance.
Many data centers have ensured responsiveness to business
requirements by over-provisioning--proactively sizing computing resources
in anticipation of peak demand. In the world of traditional servers and legacy
mainframes, over-provisioning makes sense. In fact, many advisory firms
once recommended over-provisioning as a means of meeting peak demand.

MODULAR COMPUTING

In the ideal, an alternative to over-provisioning is for IT to obtain
additional resources and bring them online as they see demand increase. In
practice, even after obtaining the hardware, setting it up and configuring the
software can take weeks. Given the real-time nature of the changes in the
computing demand, deployment takes too long, so IT began relying on over-
provisioning.
Over-provisioning has its own disadvantages. It leaves costly
resources idle most of the time. CPU utilization in many data centers range
from 15 to 20 percent for non-mainframe servers, chiefly because of
inability to rapidly reallocate unused resources during off-peak periods. Too
much capital is tied up in under-utilized resources.
To reduce capital costs, IT needs an alternative to over-
provisioning—a means of reallocating resources in minutes rather than in
weeks to accommodate peaks in demand for an application.

2.2 Cost of High Availability

As transaction processing applications have become more common,
more applications have been deemed mission-critical—capable of severely
affecting the business when they slow down or stop running altogether.
Hence the growing need for high availability.
However, high availability traditionally comes with a high price.
Redundant equipment is expensive to buy, maintain, and manage. Additional
software licenses, clustering software, and the professional services needed
to implement a traditional configuration for high availability can cost more
than the initial hardware. As a result, many IT groups continue to rely on
expensive mainframes or RISC servers that use costly switched redundant
connections to provide high availability.
Data centers need high availability, but they don’t need added
expense. They need high availability on equipments that cost less, eliminate


MODULAR COMPUTING

the need for extra software and professional services, and automate
management.

2.3 Cost of Too Many People Doing Low-Level Tasks

Labor is the largest expense associated with IT. According to Giga
Information Group, labor represents 46% of IT budgets. Finding a way to
move administrators from low level tasks to more productive tasks would
greatly improve an IT department’s ROI.

2.4 Excessive Server Management

Consider a data center with 1000 application servers. Each class of
servers has its own management and provisioning process. To support these
servers, IT needs experts for each class of server. In addition to their unique
knowledge, these experts have many redundant skills. If server management
could be simplified, many of these experts could be shifted to tasks with
higher ROI than managing servers.

2.5 Excessive Deployment Expense

Installing and configuring hardware and software takes much more
administrator time than one would expect. According to Giga Information
Group, “Management of most of large collections of servers is a manually
intensive process. Highly automated management of servers, particularly the
deployment of applications and operating systems images, is more the
exception than the rule…” Moving an application from one server to another
is a delicate task requiring days for a skilled administrator.
Complex deployment also contributes to stranded resources.

2.6 Excessive Cable Management


MODULAR COMPUTING

A full rack of traditional servers can need over 200 cables to provide
the redundant connections necessary for high availability. Such large
numbers of cables complicate cable management. Giga Information Group
says that, in large data centers that have many reconfigurations, system
administrators can spend up to 25 percent of their time managing cables.
IT needs a means of spending less time on cables.

2.7 Cost of Stranded Resources

Closely related to over-provisioning is the dilemma that causes
stranded resources. For example, suppose that demand for an application
crests, then declines over a period of months. Three factors make data center
management reluctant to harvest computing resources associated with the
application:
• The cost of the administrative time spent removing the resource from
the first application and reconfiguring it for the second.
• The risk of destabilizing the declining application.
• Removing some of the hardware used to process an application is
complex. Without extreme attention to detail, it’s possible to cause the
application to fail.
• The possibility that demand for the declining application may return
after the resources have reassigned. Should demand return, another
costly and risky harvest and reallocation would begin. The resulting
stranded resources remain unused, prematurely forcing IT groups to
buy equipment to deploy new applications and upgrade existing ones.


MODULAR COMPUTING

3 MODULAR COMPUTING ARCHITECTURE


MODULAR COMPUTING

3.1 Processing Resource Pool

It contains several diskless processing units. Modular Computing’s
processing resource is based on Intel Architecture (I A) processors because
of their superior price for performance across all business and technical
workloads. Intel server processors range from the 32-bit Intel Xeon
processor MP with strong transaction and I/O processing capabilities, to the
64-bit Itanium processor family with high performance floating point
execution. Because of their robust capabilities and price for performance,
servers based on IA processors are very popular. Processing unit refers to the
smallest chunk of processing power that can be deployed from the
processing resource pool.
3.2 Storage Resource Pool

It contains persistent storage, along with data units, operating
systems and data units. For Modular Computing, the storage resource should
be a Storage Area Network (SAN) or network-attached storage (NAS).Using
SAN or NAS allows a computing facility to concentrate the storage in one
physical location and obtain economies of scale. For example, mirroring,
backup, and offsite archiving processes are much more cost effective on a
SAN or with NAS than when applied to directly attached storage.
In addition, SAN or NAS allows the server’s personality (operating
system, application and data) to be defined completely by the content of
storage. The processing resource can be diskless and anonymous. This
allows any processing unit to be assigned to any application, facilitating the
dynamic nature of logical, rather than physical connections.

3.3 Networking Resource Pool

The networking resources should be a high-speed network accessed
through a high-speed switch. This should provide access to both the LAN


MODULAR COMPUTING

and, if needed, the internet. Just as storage resources are flexibly allocated to
meet computing demand, networking resources must be scalable so
bandwidth does not hinder performance.
3.4 Modular Computing Software (MCS)

The MCS allocates and deallocates resources from pools to virtual
servers. It also monitors virtual servers to provide band balancing and
failover. It is a vital part of Modular Computing. It obtains resources
resource pools and aggregates them into virtual servers. It also provides an
interface for administrators. Running on a control module, it can oversee
several virtual servers. The MCS also monitors the health of each virtual
server, allocates replacement, and then informs the administrator about the
status of the failed resource
3.5 Virtual Servers
Modular Computing draws elements from pools of computing
resources-processing, storage, and networking. Together, these resources
become virtual server, a computer that can be assigned to run one or more
applications. However, unlike a traditional server, when demand for an
application changes, virtual servers can dynamically be repurposed, in just
minutes.
A virtual server is logically integrated rather than physically
integrated. This distinction is essential for enabling potent management of
the resources. A control module, running Modular Computing software,
manages the creation of virtual servers and facilitates real-time allocation
and deallocation of resources.


MODULAR COMPUTING

4 ENTERPRISE MODULAR COMPUTING
Many research firms have come to the conclusion that enterprise
computing must change. Gartner Group envisions policy-based computing.
Forrester Research envisions Organic IT. Giga Information Group envisions
modularity and virtualization.
Intel is taking a leadership position in this movement. Intel expects
Modular Computing to play a major role in enterprise computing.

4.1 A New Computing Paradigm

Modular Computing relies on a new paradigm for computers.
Modular Computing draws elements from pools of computing resources-
processing, storage, and networking. Together, these resources become
virtual server, a computer that can be assigned to run one or more
applications. However, unlike a traditional server, when demand for an
application changes, virtual servers can dynamically be repurposed, in just
minutes.
A virtual server is logically integrated rather than physically
integrated. This distinction is essential for enabling potent management of
the resources. A control module, running Modular Computing software,
manages the creation of virtual servers and facilitates real-time allocation
and deallocation of resources.

4.2 Processing Resource

Modular Computing’s processing resource is based on Intel
Architecture (I A) processors because of their superior price for performance
across all business and technical workloads. Intel server processors range
from the 32-bit Intel Xeon processor MP with strong transaction and I/O
processing capabilities, to the 64-bit Itanium processor family with high
performance floating point execution. Because of their robust capabilities

MODULAR COMPUTING

and price for performance, servers based on IA processors are very popular.
Processing unit refers to the smallest chunk of processing power that can be
deployed from the processing resource pool. For example, if the processing
resource pool consists of 4-way SMPs, a processing unit is a 4-way SMP.

4.3 Storage Resource

For Modular Computing, the storage resource should be a Storage
Area Network (SAN) or network-attached storage (NAS).Using SAN or
NAS allows a computing facility to concentrate the storage in one physical
location and obtain economies of scale. For example, mirroring, backup, and
offsite archiving processes are much more cost effective on a SAN or with
NAS than when applied to directly attached storage.
In addition, SAN or NAS allows the server’s personality (operating
system, application and data) to be defined completely by the content of
storage. The processing resource can be diskless and anonymous. This
allows any processing unit to be assigned to any application, facilitating the
dynamic nature of logical, rather than physical connections. If storage were
directly attached to the processing resource, the personality would follow the
processing resource, making it less suitable for use with a different
application. If a virtual server consumes its storage resource, the Modular
Computing software automatically allocates another unit of storage to the
server.

4.4 Networking Resource

The networking resources should be a high-speed network accessed
through a high-speed switch. This should provide access to both the LAN
and, if needed, the internet. Just as storage resources are flexibly allocated to
meet computing demand, networking resources must be scalable so
bandwidth does not hinder performance.


MODULAR COMPUTING

4.5 Modular Computing Software

The Modular Computing software (MCS) is a vital part of Modular
Computing. It obtains resources resource pools and aggregates them into
virtual servers. It also provides an interface for administrators. Running on a
control module, it can oversee several virtual servers. The MCS also
monitors the health of each virtual server, allocates replacement, and then
informs the administrator about the status of the failed resource.

4.6 Fewer cables to manage

To facilitate expansion and maintenance, the processing and
networking resources, along with the control modules, could be mounted in
the same rack If this rack provides a high speed interconnect, it can reduce
the number of cables from more than two hundred to a mere handful. Two
cables from redundant switches replace the NIC cables for all servers. All
the virtual servers access the storage resource through just two cables. And a
forest of KVM cables is eliminated by providing an administrator interface
across the network.


MODULAR COMPUTING

5 BENEFITS OF MODULAR COMPUTING
Modular Computing increases agility, while reducing equipment and
people costs.

5.1 Winning through Modularity

As Janet Matsuda, SGI's director of Graphics Product Marketing,
says: "Modularity offers both savings and scalability so that customers don't
waste their money on what they don't want and can spend it on what they do
want."
Debra Goldfarb, group vice president at analyst firm IDC, agrees:
"Modular computing empowers end users to build the kind of environment
that they need not only today but over time.

5.2 Doing More With Less

To keep up with computing demand while operating within restricted
budgets, IT must find ways to optimally use computing resources and reduce
people costs. There are many areas of improvement.

5.3 Increased Agility

Changes in computing demand need no longer cause panic. The
Modular Computing software (MCS) can monitor the status of virtual
servers in real time. As demand for an application changes, the MCS can
adjust the number of virtual servers to match, in minutes instead of weeks.
This real-time load balancing prevents applications from slowing down for
long periods. The users of the applications don’t suffer lengthy response
times associated with overloaded servers.
Equipment failure no longer takes applications offline. When the
MCS detects a failure in the equipment allocated to a virtual server, the MCS


MODULAR COMPUTING

logically swaps out the failed equipment, replacing it with resources from
the pool within minutes. Applications keep running.
Because this failover capability is automatic and fast, it enables
administrators to extend failover coverage beyond mission-critical
applications to all applications running in the Modular Computing
environment.
Suppose demand grows for many applications, threatening to
regularly consume all of one of the resource pools. Rather than purchasing
an extensive, traditional server, IT purchases only the resources needed
(processing units, storage units, or network capacity) and adds them to the
resource pools. The MCS takes care of deployment in minutes as demand
fluctuates.

5.4 Reduced People Costs

With traditional, physically integrated servers, equipment failure
often means an administrator needs to visit the rack immediately to make
replacements. Each such visit is time consuming and costly. Rack visits
become rare with Modular Computing.
The Modular Computing software (MCS) acts automatically. It uses
parameters set by administrators to govern resource distribution. Once the
administrator has set the parameters, the software can balance loads or
invoke failover procedures without human intervention, in minutes.
In addition, because adding resources to a Modular Computing
environment is so easy, substantially less administrator time is spent on
configuration and setup.
Management, too, becomes easier. All applications running in a
Modular Computing environment are monitored by the Modular Computing
software. Compare this to a collection of disparate, physically integrated


MODULAR COMPUTING

servers, where each server class needs unique management tools. Reducing
the number of management tools means fewer specialized experts.
Consequently, IT management can move people from administrative
duties to activities with higher ROI, such as planning or application
development.

5.5 Reduced Equipment Costs

All applications running in a Modular Computing environment share
the same resource pools. In other words, the entire collection of virtual
servers draws load- balancing or failover resources from the same resource
pools. In contrast, with traditional computing, each mission-critical
application needs spare equipment standing by for failover or load
balancing.
With Modular Computing, a little spare resource protects all
applications. Because less resource can do the job, utilization of resources is
higher.
A related benefit of Modular Computing is the absence of stranded
resources. The MCS harvests under-utilized resources automatically.
Modular Computing helps IT do more with less. By increasing utilization of
computing resources, Modular Computing holds down capital expenditures.
By freeing administrators from tasks such as load balancing and deployment
of hardware, it makes them available for other tasks, with higher ROI.


MODULAR COMPUTING


MODULAR COMPUTING

6 APPLICATIONS

6.1 The Egenera BladeFrame system
The Egenera BladeFrame system consists of Modular Computing
software, connections for SAN or NAS and IP networking, and as many as
24 virtual servers based on Intel processors.
The BladeFrame provides a pool of up to 96 Intel processors,
deployable entirely through software, with no physical intervention. The
system components are listed in the following table.

Components and Description
Processing Blade: 2-way or 4-way, diskless, symmetric multiprocessors
(SMP5) using Intel processors. Each virtual server uses one Processing
Blade. The BladeFrame system can contain as many as 24 Processing
Blades.
Control Blade: This is the control module for the BladeFrame system. It
runs the Modular Computing software and provides security for the
Processing Blades. To ensure high availability, each BladeFrame system has
two Control Blades.
Switch Blade: This is the networking resource for the BladeFrame system. It
provides communication with the SAN or NAS and the IP network. To
ensure high availability, each BladeFrame system has two Switch Blades.
BladePlane: High-speed interconnect. Enables communication between
components within the BladeFrame system.
PAN Manager: Modular Computing software (MCS) to configure virtual
servers and govern failover and load balancing. Administrators can use the
browser-based interface or can write scripts to provide control through a
command- line interface.


MODULAR COMPUTING

The system resides in a 24x30x84-inch chassis containing a
redundant BladePlane, two Control Blades, two Switch Blades, and up to 24
Processing Blades. The BladeFrame system is a processing resource for the
data center. The Processing Blades are diskless, accessing the data center’s
storage area network (SAN) or network attached storage (NAS) for storage
resources, software, and data.
Separating processing resource from storage lets the processing
resource remain anonymous—not permanently dedicated to any particular
application(s). Anonymity facilitates reallocating Processing Blades between
the processing resource pool and virtual servers. Egenera calls this diskless
architecture a Processing Area Network, or PAN, and the management
software is called PAN Manager.
This PAN architecture facilities use of processing resources. As
demand for a particular application declines, PAN Manager software reduces
the number of virtual servers assigned to that application, making their
resources available for other applications. PAN Manager shifts resources
automatically, in minutes. By rapidly distributing resources to where they
are needed, PAN architecture eliminates costly over-provisioning. Should a
piece of equipment fail, PAN Manager detects the failure, notifies the
administrator, and allocates a replacement resource, all within minutes.
The BladeFrame system greatly reduces cable count. With traditional
architecture, each single-processor server can require more than five cables,
without providing redundancy. With the BladeFrame, as many as 96 Intel
processors can be redundantly connected to the storage and IP networks with
as few as four cables. This huge reduction in cables saves many error-prone
hours during installation, while offering fewer failure points and increased
density of servers. By reducing cable count, the BladeFrame contributes to
higher reliability (because of fewer failure points), more efficient use of


MODULAR COMPUTING

administrators (by saving cabling time), and less stranded equipment (by
simplifying harvesting and redeployment).
Many benefits of the BladeFrame system derive from the Egenera
PAN Manager software, which provides a single control point for
monitoring and allocating both physical and logical resources. Using PAN
Manager software, administrators can rapidly adjust logical configurations to
service changing demand. Tasks that were once physical and required weeks
are now accomplished through software in minutes.
The hardware and software modules of the BladeFrame system work
together to provide automation and rapid, flexible deployment. The
BladeFrame system saves administrator time associated with cable
management and other deployment issues. It automates harvesting and
reassigning resources, while slashing the cost of high availability.

6.2 NUMAflex by SGI

SGI (NYSE: SGI), known worldwide for providing a broad range of
high-performance computing and advanced graphics solutions, today
announced a technology that promises to help break the "digital ceiling"--the
performance limits that block progress in the rapidly evolving digital
economy and crucial efforts in medicine, science, manufacturing and media.
The modular, brick-style technology--called NUMAflexTM-also stands to
revolutionize the way people buy high-performance computers, allowing
them to expand and upgrade only the elements they need for their systems or
add new technologies as they become available. Traditionally, users have
had to buy expensive "one size fits all" systems that either were too much for
their needs or became obsolete quickly and had to be replaced--a costly and
cumbersome process.


MODULAR COMPUTING

"This truly is a milestone for the industry and for SGI," said Bob
Bishop, chairman and CEO, SGI. "Not only does this new technology stand
to change the way advanced computer systems are built and used, but its
flexible, cost-effective design means that more complex problems than ever
before can have access to the power of supercomputers."
The first SGI® products to utilize NUMAflex technology--the SGI®
Origin® 3000 series of servers and the SGI® Onyx® 3000 series of
visualization systems-are available immediately. A large number of orders
have been placed by notable clients such as the U.S. Army Engineering
Research Development Center and NASA/Ames Research Center. These
companies have needs for solving such demanding problems such as
financial analytics, crash-test simulation and aircraft design. In addition,
Sony Computer Entertainment Inc. has selected the SGI Origin 3400 as the
broadband server for a next-generation entertainment demonstration at
SIGGRAPH 2000.

6.2.1 About the Technology

With NUMAflex technology, each drawer-like module in a system
has a specific function and can be linked, through the patented SGI high-
speed system interconnect, to many other bricks of varying types to create a
fully customized configuration. The same bricks, depending on their number
or configuration, can be used for a continually expanding range of high-
performance computing needs: C-brick (CPU module), P-brick (PCI
expansion), D-brick (disk storage), R-brick (system/memory interconnect),
I-brick (base I/O module), X-brick (XIO expansion) and G-brick
(InfiniteReality® graphics). New brick types will be added to the
NUMAflex modular offering for specialized configurations (e.g., broadband
data streaming) and as new technologies, such as PCI-X and Infiniband,


MODULAR COMPUTING

enter the market. The systems can also be deployed in clusters or as large
shared-memory systems, depending on users' needs.
Without this modular approach, conventional high-performance
systems often need to be replaced as often as once a year to keep up with
changing needs, new technology or competitive pressures--at a cost
potentially in the millions of dollars for each replacement. This daunting
prospect can limit the progress of research and development and can hold
industries and scientific pursuits back.
"This technology represents a real revolution in thinking," said Jan
Silverman, vice president, Advanced Systems Marketing, SGI. "It's
analogous to when people switched from all-in-one stereo systems at home
to buying components for a home-based theater. Before, you had to throw
out the whole stereo because 8-track died; now you just add the DVD."

6.2.2 Executing on SGI's Corporate Strategy

From its inception, SGI has accepted the challenge of the technical
and creative user communities, working to provide them with the most
advanced computational tools. The new SGI® 3000 family is a bold and
dynamic example of the company's promise to serve these users with
industry-leading, dependable products and services that are second to none
for keeping them ahead of the technology curve and ahead of the
competition.
"SGI's customers--technical and creative computer users-are
continually demanding new products and solutions to help them reach new
heights in their own work," said Bishop. "NUMAflex modular computing is
just the latest success in our effort to meet the needs of these customers and
to help them--and SGI--stay ahead of the competition."


MODULAR COMPUTING

The new family of SGI® Origin® 3000 series servers and SGI®
Onyx® 3000 series graphics systems makes real the long-held dream of truly
modular computing. Now, technical and creative computer users can have
the same modularity, freedom of choice, and ease of upgrade that people
have long benefited from in assembling and enhancing their home-
entertainment centers. In unprecedented fashion, SGI delivers on the
promise of superior performance, custom configuration, resiliency, and
investment protection.
As Ben Passarelli, SGI's director of Server Product Marketing, says,
"With modular computing, customers can buy precisely what they need,
when they need it, with the confidence that they will be able to add the late-
breaking technologies of the future to what they already have."

6.2.3 A Superior Architecture

The newly announced SGI® 3000 family of systems marks the return
of the company to its time-honored leadership position in the realm of
technical and creative computing. The basis for the 3000 family is
NUMAflexTM technology, SGI's innovative and flexible use of a superior
supercomputer architecture.
As an architecture for high-performance multiprocessor computers,
SGI® NUMA (nonuniform memory access) exceeds the capabilities of the
SMP (symmetric multiprocessing) architecture used in previous generations
of supercomputers. SGI NUMA makes it possible for systems to increase
shared memory as needed to meet the demands of CPU-to-memory
bandwidth whenever additional processors are added to a configuration.
Growing out of a joint project with Stanford University that began more than
10 years ago, SGI NUMA gives technical and creative users superior
scalability and performance. SGI is the only computer manufacturer capable


MODULAR COMPUTING

of offering third-generation NUMA architecture, leveraging the company's
long expertise in leading-edge computing.
NUMAflex technology takes advantage of the architecture through
modular bricks that add specialized capacities in graphics, central
processing, storage, PCI expansion, or I/O capacity. Even the internal
interconnect is modular, so that large installations can be built from small
ones, one brick at a time.

6.2.4 Winning through Modularity

NUMAflex technology gives technical and creative customers
choices and growth paths never before available. As Janet Matsuda, SGI's
director of Graphics Product Marketing, says: "Modularity offers both
savings and scalability so that customers don't waste their money on what
they don't want and can spend it on what they do want."
Debra Goldfarb, group vice president at analyst firm IDC, agrees:
"Modular computing empowers end users to build the kind of environment
that they need not only today but over time. SGI, with this product, is really
ahead of the curve in the market. We are seeing the [rest of the] industry
absolutely trying to catch up" with SGI.
In addition, SGI Origin 3000 servers and SGI Onyx 3000
visualization systems reflect a return to SGI's core competencies.
"It is very exciting for us to see that SGI is once again really
becoming true to the mission it had years ago, that of leading the industry in
technical computing, " says Goldfarb. "This company has really hit it this
time and [we] believe this is really the right technology at the right point in
time."


MODULAR COMPUTING

6.2.5 The Power of Visualization

Of course visualization, along with data handling and scalable
architecture, has always been one of SGI's three main core competencies.
The new SGI Onyx 3000 series, which utilizes next-generation
InfiniteReality3TM graphics, will be able to aid users in what Matsuda calls
"their need to understand." Says Matsuda, "You can get powerful
visualization with powerful computing, because your eyes are the widest
channel to the brain. And sometimes you need to give people experiences
you don't want them to have in real life."
A unique feature of InfiniteReality3 is its ability to perform visual
serving, delivering powerful graphics capabilities over a network as needed.
The new SGI Onyx 3000 series systems are also optimized for real-time
simulation, such as in planetariums, Reality Center® facilities, digital media
and geospatial imaging.
A final component of SGI's renewed focus on its customers and what
Passarelli calls "working to our strengths" is SGI's continuing strong
commitment to both MIPS® and IRIX®, which is evidenced by
unprecedented customer demand for the new product line. While SGI sees
long-term strategic value in the company's involvement with the Open
Source community, "We remain fanatically committed to helping our
customers solve their problems in the here and now. For customers on the
leading edge, if you give them more capabilities, more compute power, and
greater visualization, they can do amazing things."


MODULAR COMPUTING

7 REMARKS FROM EXPERTS

Call it grid computing. Or modular computing. Or policy-based
computing or utility computing. Intel, which is opting for the modular
designation, is preaching distribution of processing power to boost
performance and reliability. Modular computing represents a new paradigm
that requires advances in both software and hardware, according to Intel.
"There (are) a lot of people that associate modular computing (with) blades
and blade form factors. It's important to know this is far more than form
factors and far more than blades," said Abbi Talwalkar, vice president of the
Intel platform products group, in Hillsboro, Ore., during a presentation at the
Intel Developer Forum.
Modular computing, the joining of multiple computing resources, is
an answer for exponential data growth, application and server sprawl, and
dis- aggregation of storage, according to Intel. The concept also is critical in
today's tough economic times, with IT cutbacks, Talwalkar said. Modular
computing is characterized by a growth in hardware clustering and
distributed computing along with software developments such as the
deployment of application servers and the use of Web services for
intersystem communication, he said. "It's really advances in system
management and clustering technology that's going to drive much of the
adoption here," Talwalkar said. Clustering might displace large symmetric
multiprocessing systems over time, he said. Automation, enabling for
dynamic allocation of resources, is probably the "heart" of modular
computing, according to Talwalkar. Automation developments are needed
such as self-healing systems, failover, and dynamic performance
optimization, he said.
Benefits of modular computing include maximization, efficiency,
Internet reliability, and seamless and simplified management, according to
the company. For example, modular computing will maximize use of a

MODULAR COMPUTING

server that might have 40 percent of its capacity not being used, Talwalkar
said. "Software is going to drive the success of modular computing 100
percent," Talwalker stressed.
One IDF attendee, however, criticized Intel for recently backing away
from plans to produce InfiniBand-based hardware. InfiniBand, said Anil
Vasudeva, president and CEO of research firm Imex Research, of San Jose,
is key to making blade servers function together. InfiniBand is a next-
generation switched-fabric I/O technology. "Intel seems to have done a big
boo boo job on that," Vasudeva said. Talwalkar said that given current
economics, there were "some very difficult decisions to make at Intel in
terms of productizing components."


MODULAR COMPUTING

8 CONCLUSION

Stretching the IT Dollar

Modular Computing replaces the physical connections between
computing resources with logical connections. Because the connections are
logical. Modular Computing software can monitor and control how virtual
servers use resources.
This software-based monitoring and controlling enables automated
resource management, where the software continuously redistributes
resources according to parameters provided by an administrator. This, along
with simplified server management and reduced cable count, means large
collections of servers need fewer administrators.
Modular Computing uses small amounts of spare resource to provide
failover and load balancing for all applications running in the Modular
Computing environment. This eliminates stranded resources, boosts resource
utilization, and holds down capital expense.
Because Modular Computing is built on IA processors, it offers better
price for performance and ensures a broader choice of software vendors and
software. A larger selection of software can speed application development,
and competition between software vendors can hold down development
costs.
Modular Computing is a concept for the future, hut it is available
now, in products shipping today. It is already proving itself by saving money
for IT.


MODULAR COMPUTING

REFERENCES

• Forrester Research, Inc., The New Computing Utility.
• Goldman Sachs, IT Spending Survey.
• Giga Information Group, Inc., The Future of the Data Center-
Modularity and Virtualization.
• Gartner, Inc., Budgeting for IT-Average Spending Report.
• www.pcmag.com
• www.pcquest.com
• www.itpapers.com


Modular Computing

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Modular Computing (20)

Recently uploaded (20)

Modular Computing