Project DRAC: Creating an applications-aware network
1 like339 views
Nortel has developed the Dynamic Resource Allocation Controller (DRAC), a proof-of-concept software that enhances network resource management by enabling applications to directly control their shares of network resources without operator involvement. DRAC addresses challenges associated with diverse network technologies and manual provisioning, improving efficiency and reducing operational costs in various fields such as healthcare and grid computing. Demonstrated in real-world environments like Surfnet's hybrid optical and packet network, DRAC has shown significant benefits in user satisfaction and resource optimization.
![What is SURFnet?
SURFnet is a national high-performance
computer network in the Netherlands that
connects more than 150 institutions in
higher education and research to each
other, as well as to other networks
around the world.
Operated by a Netherlands-based
company, SURFnet is among the leading
research networks in the world, collabo-rating
closely with national and interna-tional
organizations and serving as an
advanced test environment for investigat-ing
new technologies that will continually
improve the reliability, security, and
Page 26
speed of its network.
Among its many ongoing projects,
SURFnet is responsible for the
realization of the GigaPort Next-
Generation Network – a project of the
Dutch government, trade and industry,
and educational and research institu-tions
– that aims to strengthen the
national knowledge infrastructure.
Research on optical and IP networking
and grids is a prominent part of the
project. For more information, visit
http://www.surfnet.nl/.
tan area networks (MANs) when
operating replication, business
continuance, and disaster recovery
applications. During trials completed
with some of Nortel’s storage
partners, the control and monitoring
of storage and network management
functions were consolidated into
one unified “cockpit” versus two, to
command and control these func-tions,
with DRAC discovering the
topology and composing a complete
system view inclusive of the storage
topology.
• Healthcare workflows (such as in
radiology practices) are a natural fit
for the DRAC’s ambitions in WENs.
Beyond the bandwidth savings seen
in data center scenarios, the WEN
can improve on dependability, while
optimizing the expenses in network
and storage setups. [For further
detail, refer to Schofield’s paper
“Workflow Engaged Networks for
Radiology in Metro Regions,”
presented at the 90th RSNA, Radio-logical
Society of North America,
Chicago, November 28, 2004. This is
the study of a network (wireline,
LAN+MAN) utilized for Filmless
Radiology, with and without a DRAC
framework.]
• Within grid computing communi-ties,
we are working to elevate the
network to a primary grid-managed
resource, akin to CPU and storage
resources. DRAC can tame the
complexity and diversity of network
elements to open the way for
e-utilities.
Demonstrating the values
DRAC has demonstrated compelling
values in a wide range of applica-tions.
Detailed studies by Nortel have
shown large cost savings in MANs
and wide area networks (WANs),
providing an appealing alternative to
the old approach of static, over-provisioned
networks.
As well, DRAC is undergoing
validation in real-world network
deployments with high-performance
computing networks. For instance,
the Netherlands-based SURFnet is
currently deploying DRAC at the
heart of a hybrid optical and packet
network, called SURFnet6, which is
being realized in the context of the
GigaPort Next-Generation Network
project (see sidebar). (SURFnet 6
optimizes Layer 1 network resource
utilization based upon end-user
requirements.) In this implementa-tion,
applications driving multi-Gbit/s
transfers bypass the packet layer and
are steered directly onto wavelengths
between end points across tempo-rarily
assigned optical links.
In addition to the SURFnet project,
Nortel recently demon-strated
DRAC features at
the Supercomputing 2004
conference in November
2004. This demonstration
was done in cooperation
with recognized research
and education leaders,
including SURFnet,
Netherlight, the Univer-sity
of Amsterdam,
Internet2, Canarie, iCAIR,
and Starlight. Encouraged
by the results of the DRAC
preliminary results,
Nortel is currently look-ing
to a broader deploy-ment
market opportunity.
Franco Travostino is leader of an
Advanced Technology Team that is
exploring applications-engaged
networks and grid infrastructures.
This team includes Tal Lavian, Inder
Monga, and Bruce Schofield.
Rob Keates is senior manager,
optical networks marketing.
Note: This article has been adapted
from a Nortel Applications Brief
(www.nortel.com/drac), published
in November, 2004, and distributed
at Supercomputing 2004, held in
Pittsburgh, Pennsylvania.](https://image.slidesharecdn.com/projectdraccreatinganapplications-awarenetwork-141113111316-conversion-gate01/85/Project-DRAC-Creating-an-applications-aware-network-4-320.jpg)
Project DRAC: Creating an applications-aware network
- 1. by Franco Travostino, Rob Keates, Tal Lavian, Inder Monga, and Bruce Schofield Across a number of different fields – health care, manufacturing, and research and education, for example – users are demanding more and more network resources, such as bandwidth, quality of service, and security, in order to collaborate and share data around the globe. To meet these demands, the CIOs for these institutional, commercial, and research pools have employed several solutions, including supercomputing technologies, grid computing, and peak provisioning and manual provisioning of network resources. None of these methods, however, is optimal. Supercomputing technology, for example, is expensive to scale and limited to those institutions and researchers that can afford it; high-performance grid computing, while it enables multiple users to share resources to boost processing power, is difficult to achieve due to the cost of interconnecting proces-sors with low latency and high bandwidth; and statically and peak-provisioned network setups can result in over-provisioning of network resources, which is expen-sive and results in CapEx and OpEx inefficiencies. At the same time, operators must optimize their networks to meet diverse user requirements, which can range from single-user multi- Gbit/s data transfers (as with grid applications), to best-effort many-to-many kilobytes (such as e-mail). How can the network provide the cost/performance profile of lower-layer (e.g. optical) technologies while achieving the dynamic capabilities of higher layers? As well, how can the network adapt to the evolving applications? Manual provisioning of network resources has also been known to present challenges. For example, processing their massive data files requires researchers to first contact a network administrator to set up and provision the appropriate network resources – typically a manual, error-prone task accomplished through a point-and-click session operated by the network administra-tor, which can lead to delays and potential failures. Moreover, as data travels through the end-to-end network and across different networks, it typically encounters different types of network technologies – from packet, circuit, wireless, and wireline to various access environments – each with its own separate topologies, protocols, and features, again leading to missed opportunities or high CapEx/OpEx costs. Dynamic Resource Allocation Controller To address these challenges, Nortel has developed a proof-of-concept capability, called Dynamic Resource Allocation Controller (DRAC, pronounced d-rack). Essentially, DRAC acts as an agent of the various applications, brokering and configuring on an end-to-end basis all the necessary pieces of the network, regardless of the type of network – circuit or packet, wireless or wireline. DRAC enables applications to control their share of network resources, yet without requiring them to interface directly with a wide range of diverse and constantly evolving network protocols, features, and devices. Put another way, DRAC lays the tracks ahead of the train, adjusts the network resources that an applica-tion needs, and steers the data through the network – and it does this either dynamically in real time or on a time reservation basis. DRAC is implemented as software that is designed to be portable to any Java platform. This middleware sits between applications and the network (whether management Page 23 Project DRAC: Creating an applications-aware network Intelligent networking and the ability for applications to more effectively use all of the network’s capability, rather than just the transport “pipe,” have been elusive. Until now. Nortel has developed a proof-of-concept software capability — service-mediation “middleware” called the Dynamic Resource Allocation Controller (DRAC) — that runs on any Java platform and opens up the network to applications with proper credentials, making available all of the properties of a converged network, including service topology, time-of-day reservations, and interdomain connectivity options. With a more open network, applications can directly provision and invoke services, with no need for operator involvement or point-and-click sessions. In its first real-world demonstrations in large research networks, DRAC is showing it can improve user satisfaction while reducing network operations and investment costs.
- 2. planes, control planes, or individual network elements) and provides applications with an abstracted view of the underlying network (Figure 1). The DRAC southbound interface features a framework for the instantiation of multiple “drivers,” corresponding to the different signaling techniques encountered in legacy networks (Figure 2). Three degrees of coupling With the DRAC concept, Nortel envisions three degrees of coupling between applications and networks. • First degree: In hybrid optical and packet networks, such as SURFnet6 (discussed later in this article), DRAC provides “cut-through” capabilities across network layers by steering very large flows of packets or low-latency applications dynami-cally over Layer 1 instead of Layer 3. For example, instead of dedicating routing resources to multi-Gigabit point-to-point file sharing applica-tions or alternatively setting up a dedicated and costly high-band-width optical connection, DRAC simply sets up and takes down “ephemeral” optical circuits as they are needed, minute-by-minute, hour-by-hour. By bypassing the routing layer for this type of traffic, DRAC enables a higher-performance network experience for both routed and bypassed traffic and reduces the total number of routers required in the network. In fact, in one real-world design, DRAC reduced the number of required routers from 20 to 2. The same thesis can be applied to other environments featuring diverse technologies, such as between wireline and wireless. • Medium degree: DRAC is capable of recognizing the network foot-prints of a given application (through deep packet inspection or direct signaling from the application, for example). DRAC makes sure that the network reacts appropriately to an application’s behavior. For Page 24 Figure 1. DRAC core framework Grid applications: financial (e.g. stats analysis); manufacturing (e.g. CAD); entertainment (e.g. digital rendering) Applications: Business process workflow, grid resource manager, storage, video streaming, converged services Multiple applications DRAC: service-mediation software Multiple networks Applications access network services Receive network feedback • Fault notification • Abstracted network and performance view • Application request Application-facing API • AAA services • Abstraction • Reservation, workflows • Smart bandwidth management • Virtualization topology Internal APIs policy Provisioning and signaling protocols Network resources The DRAC core framework includes AAA (authentication, authorization, accounting) services, policy engine, topology discovery engine, resource reservation services, workflow utilities, interdomain routing facilities, and smart bandwidth management fixtures. The interface to applications is bi-directional, enabling network performance and availability information to be abstracted upward toward the application. The DRAC provides applications with the means to directly drive their share of network resources within a policy-defined envelope of flexibility. Network resources include bandwidth, quality of service (QoS), security, acceleration appliances, sensors, and more. As well, the DRAC strategy is to use existing standards and toolsets for interfaces, which greatly simplifies deployment in multivendor, multi-technology environments. • Web Interface • Legacy IP/QoS • ASTN UNI • (G)MPLS • CIM • SNMP • TL1 • Others
- 3. example, when a critical storage restore operation is initiated due to disaster recovery, DRAC ensures that the network dedicates a large fraction of its resources to expedite that operation. • High degree: DRAC becomes privy to the overall “flight plan” of an application. For instance, DRAC learns how a particular workflow unfolds among peering instances of a distributed application. That way, DRAC can anticipate the network requirements, evaluate what-if scenarios, and enact failure-recovery strategies that are cognizant of the workflow. These are the defining properties of what we call “workflow-engaged networks” (WENs). In all cases, DRAC enables much more efficient use of network resources, leading to operational and capital savings. Application value Currently, the DRAC value proposi-tions have been validated within four vertical market segments: • Hybrid optical and packet net-works validate the DRAC cut-through capabilities by steering very large data flows across an ephem-eral optical circuit, and allowing smaller flows that are more tolerant to latency and/or congestion loss to communicate via Layer 3. • Within data centers, DRAC will help storage partners realize large bandwidth savings across metropoli- Page 25 Multiple applications Figure 2. DRAC middleware Multiple applications OAM OAM OAM OAM OAM OAM DRAC sits between applications and the various networks that connect end systems, such as servers, storage, visualization devices, and sensors. Value-add services Session establishment end-to-end Legacy sessions (management and control planes) Core networks Metro networks Access networks 3rd-party services Grid community scheduler Workflow language DRAC DRAC DRAC DRAC DRAC Control plane B Control plane A Servers Storage Visualization Sensors OAM OAM OAM Servers Storage Visualization Sensors
- 4. What is SURFnet? SURFnet is a national high-performance computer network in the Netherlands that connects more than 150 institutions in higher education and research to each other, as well as to other networks around the world. Operated by a Netherlands-based company, SURFnet is among the leading research networks in the world, collabo-rating closely with national and interna-tional organizations and serving as an advanced test environment for investigat-ing new technologies that will continually improve the reliability, security, and Page 26 speed of its network. Among its many ongoing projects, SURFnet is responsible for the realization of the GigaPort Next- Generation Network – a project of the Dutch government, trade and industry, and educational and research institu-tions – that aims to strengthen the national knowledge infrastructure. Research on optical and IP networking and grids is a prominent part of the project. For more information, visit http://www.surfnet.nl/. tan area networks (MANs) when operating replication, business continuance, and disaster recovery applications. During trials completed with some of Nortel’s storage partners, the control and monitoring of storage and network management functions were consolidated into one unified “cockpit” versus two, to command and control these func-tions, with DRAC discovering the topology and composing a complete system view inclusive of the storage topology. • Healthcare workflows (such as in radiology practices) are a natural fit for the DRAC’s ambitions in WENs. Beyond the bandwidth savings seen in data center scenarios, the WEN can improve on dependability, while optimizing the expenses in network and storage setups. [For further detail, refer to Schofield’s paper “Workflow Engaged Networks for Radiology in Metro Regions,” presented at the 90th RSNA, Radio-logical Society of North America, Chicago, November 28, 2004. This is the study of a network (wireline, LAN+MAN) utilized for Filmless Radiology, with and without a DRAC framework.] • Within grid computing communi-ties, we are working to elevate the network to a primary grid-managed resource, akin to CPU and storage resources. DRAC can tame the complexity and diversity of network elements to open the way for e-utilities. Demonstrating the values DRAC has demonstrated compelling values in a wide range of applica-tions. Detailed studies by Nortel have shown large cost savings in MANs and wide area networks (WANs), providing an appealing alternative to the old approach of static, over-provisioned networks. As well, DRAC is undergoing validation in real-world network deployments with high-performance computing networks. For instance, the Netherlands-based SURFnet is currently deploying DRAC at the heart of a hybrid optical and packet network, called SURFnet6, which is being realized in the context of the GigaPort Next-Generation Network project (see sidebar). (SURFnet 6 optimizes Layer 1 network resource utilization based upon end-user requirements.) In this implementa-tion, applications driving multi-Gbit/s transfers bypass the packet layer and are steered directly onto wavelengths between end points across tempo-rarily assigned optical links. In addition to the SURFnet project, Nortel recently demon-strated DRAC features at the Supercomputing 2004 conference in November 2004. This demonstration was done in cooperation with recognized research and education leaders, including SURFnet, Netherlight, the Univer-sity of Amsterdam, Internet2, Canarie, iCAIR, and Starlight. Encouraged by the results of the DRAC preliminary results, Nortel is currently look-ing to a broader deploy-ment market opportunity. Franco Travostino is leader of an Advanced Technology Team that is exploring applications-engaged networks and grid infrastructures. This team includes Tal Lavian, Inder Monga, and Bruce Schofield. Rob Keates is senior manager, optical networks marketing. Note: This article has been adapted from a Nortel Applications Brief (www.nortel.com/drac), published in November, 2004, and distributed at Supercomputing 2004, held in Pittsburgh, Pennsylvania.