Limiting Latencies in the Cloud

By Dr. Casimer DeCusatis

October 22, 2009

One of the primary reasons for the phenomenal interest today in cloud computing is that it abstracts technological complexity away from the end user. Innovative high-bandwidth, low-latency services are accessed from the cloud, without much worry about the hardware or software resources deployed locally.

For the operators of networks that underlie cloud computing, of course, the technological considerations are not small. Whether engineering a private cloud to be operated by an enterprise information technology (IT) staff for that enterprise’s users or a public cloud to be operated by a provider of managed services to business customers, one of the key decisions to be made is how to interconnect servers and storage within and among datacenters and create virtual machines.

There are a number of technology options available to IT managers and service planners who need high-bandwidth, low-latency transport for virtualized sharing of resources across the cloud. Interconnect latencies have been generally recognized as a limiting factor for high-performance computing applications in and among cloud centers, but a variety of protocol innovations have appeared in the marketplace to clear this hurdle.

Behind the Fervor

The traditional datacenter compute model has been characterized by lightly utilized x86-architecture servers running one bare-metal operating system or multiple operating systems via a hypervisor. Lower-bandwidth links among server resources have been sufficient, given the low number of virtual machines in use.

Today, industry is moving full-bore toward a more dynamic model. In cloud computing, servers are more highly utilized and clustered to support many virtual machines. Cloud implementations — in which software-based services are decoupled from particular servers and other hardware resources — vary widely.

Internet search engines and social media sites have leveraged this cloud computing approach for some time. The trend has extended to traditional business services and applications, as evidenced with the appearance of Google Docs and Salesforce.com, for example. Public desktop cloud service has emerged, in which end users access applications and data from network-attached PCs or other devices. Even high-performance computing is likely to leverage the cloud.

The business benefits are considerable. Reductions are to be realized in both capital and operating expenditures (CAPEX and OPEX). Higher utilization of mid-range servers, storage and converged network fabrics results in reduced CAPEX. Automated, integrated management of the end-to-end datacenter infrastructure lowers OPEX.

Interconnect Requirements

Some of the cloud computing applications, such as the recently emerged desktop capability, might allow for latencies of 50 milliseconds or more. Others, such as in high-performance computing, will demand ultra-low latency near one millisecond. A number of protocols vie for the role of datacenter interconnect for these more demanding applications.

Business-continuity and disaster-recovery solutions based on IBM’s Geographically Dispersed Parallel Sysplex (GDPS), for example, rely on screaming-fast InfiniBand for interconnection among remotely located datacenters and links to virtual storage.

An InfiniBand port — supporting up to 40Gb/s of bandwidth and as little as 1 microsecond of latency — is necessary for the highest-performance, transaction-sensitive applications. (In addition, InfiniBand cuts the I/O overhead introduced by virtualization in cloud computing.) An Ethernet port, by comparison, offers 10Gb/s of bandwidth and 6 microseconds of TCP protocol latency — though latencies as great as 40 to 50 microseconds are not uncommon. This is because larger networks interconnected via Ethernet ports sometimes require multiple tiers of hierarchical switching to offset the impact of oversubscription.

As for interconnection to storage resources within the datacenter, 8Gb/s Fibre Channel is the dominant protocol for large enterprise applications — enabling rapid backup and recovery when transported natively via Dense Wavelength Division Multiplexing (DWDM). It will need to be accommodated in the cloud, where storage requirements are tremendous.

The maturity of a protocol is a key consideration, as proven technologies will be a must for the mission-critical, high-bandwidth, low-latency applications now being supported via clustered virtual machines. For the foreseeable future, proven implementations of each of these protocols — InfiniBand, 8G Fibre Channel and some form of Ethernet — are likely to have roles in the cloud, because enterprise IT managers and carrier service planners are generally unwilling to commit their cloud infrastructures to any one of them alone.

DWDM accommodates high-performance, low-latency transport of each of these protocols in their native form over fiber spans of up to 600 kilometers. There is no need for the operator of a cloud computing network to take on the costs and complexities of installing channel-extender gateways. Plus, protocol-agnostic DWDM offers operators of cloud computing networks the flexibility to adopt emerging protocols — such as promising Fibre Channel over Ethernet (FCoE) — as they mature.

Convergence Hopes

Eventually, the future cloud datacenter is likely to rely on some form of single, converged fabric with server and storage virtualization. This vision — one flexible, reliable, high-performance protocol cost-effectively undergirding all local and storage area network (LAN and SAN) traffic — has been espoused in enterprise networking for years, and operators of cloud-computing infrastructures will seek to reduce the cost of cloud bandwidth as long as the required performance characteristics can be dependably delivered.

There are technologies on the horizon that are being touted for this convergence role. For example, interest in various forms of InfiniBand over Ethernet (IBoE) is gathering, but industry standards are a long way away. FCoE also bears watching.

In its current form, FCoE offers valuable I/O consolidation for racks of single-rack-unit or blade servers running new converged network adapters (CNAs). Emerging FCoE switches behave effectively as top-of-the-rack aggregators, distributing data traffic to either legacy LAN or SAN infrastructure. The performance characteristics promised by this technology — low latency and 10 to 40Gb/s bandwidth — are intriguing.

The enthusiasm for FCoE among operators of cloud-computing infrastructures is tempered by several factors, though. First, the protocol is far from proven in the kind of demanding, large-scale deployment that is typical of high-performance computing. Second, there are multiple issues with FCoE in relation to distance:

  • No existing standard yet defines the implementation of Inter-switch Links (ISLs) among geographically dispersed FCoE switches.
     
  • Multi-hop support is still forthcoming.
     
  • The ability to natively connect the technology across 100 kilometres or more has not been demonstrated.
     
  • Very few storage vendors offer native FCoE interfaces.

Until these enhancements come about, FCoE will play a limited role in the cloud.

Conclusions

While there is a clear desire to ultimately converge all enterprise traffic on a single unifying fabric, the real-world cloud-computing environment based on high-end servers must continue to support a range of existing multiprotocol fabrics that are optimized for a range of disparate tasks. InfiniBand, for example, provides the extremely high bandwidth and extremely low latency required by the most demanding of the widening array of cloud applications, such as those found in high-performance computing.

Enterprise IT managers and carrier service planners must continue to be savvy about matching technologies with applications to most cost-efficiently meet technical requirements and business objectives. InfiniBand, 8G Fibre Channel, flavors of Ethernet and other protocols such as iSCSI are likely to play important roles in enterprise networking for the next several years, and network operators who are introducing cloud computing must plan accordingly. Public cloud service providers, for example, should consider the potential benefits of direct attach storage into the cloud as an alternative to deploying and managing large numbers of FC-IP gateways.

The other problem with FC/IP gateways is the fact that storage performance is hamstrung if, for example, a 2G Fibre Channel stream is compressed to Gigabit Ethernet or OC-48s when native 4G, 8G, 10G and emerging 16G Fibre Channel are options at the same distances.

DWDM enables the full gamut of protocols to be commonly deployed and managed across existing optical networks. Because it enables each of the interconnect protocols to perform seamlessly at wire speed, without introducing additional latency, WDM is well positioned to deliver the unifying function network operators need as they roll out cloud computing services.

About the Authors

Dr. Casimer DeCusatis is an IBM Distinguished Engineer based in Poughkeepsie, NY, where he currently serves as Chief Engineer of the IBM/Juniper alliance and architect for network and I/O solutions, including extended distance connectivity.  He is an IBM Master Inventor with over 85 patents, and recipient of several industry awards, including the IEEE Kiyo Tomiyasu Award, the EDN Innovator of the Year Award, the Mensa Research Foundation Copper Black Award for Creative Achievement, and the IEEE/HKN Outstanding Young Electrical Engineer award (including a citation from the President of the United States and an American flag flown in his honor over the U.S. Capitol). He is co-author of more than 100 technical papers, book chapters, and encyclopedia articles, and editor of the Handbook of Fiber Optic Data Communication (now in its 3rd edition). He is a member of the IBM Academy of Technology and co-leader of the Academy study “Innovation Ecosystems.” Dr. DeCusatis received the M.S. and Ph.D. degrees from Rensselaer Polytechnic Institute (Troy, NY) in 1988 and 1990, respectively, and the B.S. degree magna cum laude in the Engineering Science Honors Program from the Pennsylvania State University (University Park, Pa.) in 1986. He is a Fellow of the IEEE, Optical Society of America, and SPIE (the international optical engineering society), a member of the Order of the Engineer, Tau Beta Pi, Eta Kappa Nu, Mensa, and various other professional organizations and honor societies.  He also serves as Founder and Director of Hudson Valley FIRST Lego League, which offers over 1,000 students each year the opportunity to pursue their interest in science and technology.

Todd Bundy is Director Global Alliance Management Enterprise with ADVA Optical Networking. He has 26 years of experience in the storage networking industry, is a recognized expert in SAN and optical networking, and specializes in storage applications over various types of networks to meet corporate contingency plans. Mr. Bundy has participated in many successful large-scale Disaster Recovery and Data Center Consolidation projects with companies like IBM, using ADVA Optical Networking FSP (Fiber Service Platform) WDMs.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, code-named Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from its predecessors, including the red-hot H100 and A100 GPUs. Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. While Nvidia may not spring to mind when thinking of the quant Read more…

2024 Winter Classic: Meet the HPE Mentors

March 18, 2024

The latest installment of the 2024 Winter Classic Studio Update Show features our interview with the HPE mentor team who introduced our student teams to the joys (and potential sorrows) of the HPL (LINPACK) and accompany Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the field was normalized for boys in 1969 when the Apollo 11 missi Read more…

Apple Buys DarwinAI Deepening its AI Push According to Report

March 14, 2024

Apple has purchased Canadian AI startup DarwinAI according to a Bloomberg report today. Apparently the deal was done early this year but still hasn’t been publicly announced according to the report. Apple is preparing Read more…

Survey of Rapid Training Methods for Neural Networks

March 14, 2024

Artificial neural networks are computing systems with interconnected layers that process and learn from data. During training, neural networks utilize optimization algorithms to iteratively refine their parameters until Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, code-named Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi Read more…

Survey of Rapid Training Methods for Neural Networks

March 14, 2024

Artificial neural networks are computing systems with interconnected layers that process and learn from data. During training, neural networks utilize optimizat Read more…

PASQAL Issues Roadmap to 10,000 Qubits in 2026 and Fault Tolerance in 2028

March 13, 2024

Paris-based PASQAL, a developer of neutral atom-based quantum computers, yesterday issued a roadmap for delivering systems with 10,000 physical qubits in 2026 a Read more…

India Is an AI Powerhouse Waiting to Happen, but Challenges Await

March 12, 2024

The Indian government is pushing full speed ahead to make the country an attractive technology base, especially in the hot fields of AI and semiconductors, but Read more…

Charles Tahan Exits National Quantum Coordination Office

March 12, 2024

(March 1, 2024) My first official day at the White House Office of Science and Technology Policy (OSTP) was June 15, 2020, during the depths of the COVID-19 loc Read more…

AI Bias In the Spotlight On International Women’s Day

March 11, 2024

What impact does AI bias have on women and girls? What can people do to increase female participation in the AI field? These are some of the questions the tech Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Analyst Panel Says Take the Quantum Computing Plunge Now…

November 27, 2023

Should you start exploring quantum computing? Yes, said a panel of analysts convened at Tabor Communications HPC and AI on Wall Street conference earlier this y Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Training of 1-Trillion Parameter Scientific AI Begins

November 13, 2023

A US national lab has started training a massive AI brain that could ultimately become the must-have computing resource for scientific researchers. Argonne N Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire