At 100 Gbps, ESnet Puts Network Research on Fast Track

By Nicole Hemsoth

September 12, 2012

When the Department of Energy (DOE) announced the Advanced Networking Initiative in 2009 to develop the first 100 gigabit-per-second (Gbps) production-ready science network, it also included funding for a 100 Gbps experimental testbed and a national dark-fiber testbed. The thinking was that researchers from government institutions, universities and industry could use these testbeds to experiment with disruptive network technologies without interfering with traffic on a major production network.

The 100 Gbps testbed provides a facility for researchers to solve the near-term challenges associated with deploying and operating these types of networks, while the dark fiber testbed allows for research into disruptive technologies and approaches that are still not ready for production use.

The testbeds are operated by DOE’s ESnet, the Energy Sciences Network, which is managed by Lawrence Berkeley National Laboratory.

So far about 20 groups have taken advantage of ESnet’s 100 Gbps testbed, which currently connects two DOE supercomputing facilities at 100 Gbps: the National Energy Research Scientific Computing Center (NERSC) in Oakland, Calif., and the Argonne Leadership Computing Facility (ALCF) in Argonne, Ill. Additional connections will include the Oak Ridge Leadership Computing Facility (OLCF) in Oak Ridge, Tenn., at 80 Gbps, and Fermilab in Batavia, Ill., at 50 Gbps.

Recently, the testbed received additional funding for continued operations, providing a unique resource for experimenting with potentially disruptive network technologies. Brian Tierney, who leads ESnet’s Advanced Networking Technologies Group, talks more about the project in this Q&A with Jon Bashor of Berkeley Lab Computing Sciences Communications.

First, can you give us an overview of the testbed, who gets to use it and why it’s important?

Brian Tierney: Originally part of the now-completed Advanced Networking Initiative (funded by the Recovery Act), the testbed features a set of fast end hosts at NERSC and Argonne, with a dedicated 100 Gbps connection linking them. When a researcher is allocated time, his or her project is the only traffic on the testbed, allowing them to run experiments in a controlled environment. One of the challenges in network research is repeatability, so giving a researcher complete control of a 100 Gbps testbed allows the experiment to be re-run multiple times, enabling them to adjust the experiment if needed, leading to more exact results.

The testbed is open to industry, government labs and academia to use, and we accept proposals twice a year. The next proposal deadline is Monday, Oct. 1. It’s a fairly simple proposal process:  just a two to three page description of your proposed research and why you can’t do this research elsewhere.  Proposals are reviewed by an external committee with representatives from academia and industry. So far we’ve had a good mix of high quality projects, some with DOE funding, others are NSF-funded, others from NASA and industry.

Why is 100G research important?

Tierney: Just because the network is 10 times faster does not mean the protocols and middleware will be 10 times faster. Various host-tuning parameters are different, and new data movement protocols or middleware design may be needed.

What’s DOE’s interest in this work?

Tierney: You’ve probably heard about the President’s Big Data Initiative, which was announced in June. Many sources of Big Data are facilities at DOE laboratories, such as the Advance Light Source here at Berkeley Lab, the Advanced Photon Source at Argonne, and the Spallation Neutron Source at Oak Ridge. Then there are experiments like those at the Large Hadron Collider (LHC) in Switzerland, in which DOE is a very active collaborator.

With more experimental facilities coming online and others receiving major upgrades, the datasets are only getting bigger and bigger, which makes them harder to manage and share with other researchers. So DOE is highly interested in exploring new protocols for moving these big datasets, some of which are being explored on the testbed.

What’s been the response from the community?

Tierney: It’s been extremely popular. We officially opened the testbed in January 2012, and it’s been booked almost 24×7 since then. There are some grad students out there who stay up very late sometimes!  So far, we’ve relied on sending announcements to several email lists, presentations at various conferences and workshops, and via our collaborator organizations, as well as working through program managers at DOE and the National Science Foundation for submissions.

Our current policy is to allow each project up to one six-hour slot per day. Six or seven projects are actively using the testbed, and five or six more have been granted access and are now gearing up to begin testing.

One of the most active testbed projects is from Fermilab, one of two Tier 1 sites for LHC data in the U.S. Researchers there are trying to develop innovative middleware for moving large datasets more efficiently. Experiments by staff at NERSC and the leadership computing centers at Argonne and Oak Ridge have also been very active. A number of interesting industry collaborations are using the testbed as well.

What do you think is the reason for this strong response?

Tierney: For the networking research community, there is no other test environment like this that provides researchers the ability to experiment with their ideas “at scale” — which means they don’t have to rely on a local area test loop to prove an idea that is meant to be used on a national backbone. You really need a cross-country environment to test those ideas. And 100 Gbps technology is expensive — none of the research groups could afford to build an environment like this on their own. We’re providing access at no cost.

To my knowledge, there is nothing else out there at 100 Gbps. In addition, we are providing “bare metal” access to the hosts, not just virtual machines. Before each project’s test slot, we boot up that user’s own image on the hosts. We’ve gotten a lot of positive feedback about this model.

Because this is a research testbed, we are strongly encouraging the users to publish their results, and sometimes collaborate with research groups on their papers. One of those, “Protocols for Wide-Area Data-Intensive Applications — Design and Performance Issues” by Yufei Ren, Tan Li, Dantong Yu, Shudong Jin, Thomas Robertazzi, Eric Pouyoul and myself, will be presented at SC12 in Salt Lake City.

At Berkeley Lab, we’ve been involved with testbeds since the early 1990s. Back then there was BAGNet, the Bay Area Gigabit Network — that’s one gigabit. I also worked on the MAGIC Gigabit Testbed, which tested some of the early ATM (asynchronous transfer mode) equipment. And we are currently talking to other testbed projects, including NSF’s GENI, an academic testbed for software-defined networking, and OFELIA, an OpenFlow experimental environment in Europe.

Can you give a few examples of what you think have been the coolest projects?

Tierney: There have been three separate projects testing RDMA (Remote Direct Memory Access) Over Converged Ethernet, or RoCE. This is a protocol that bypasses traditional TCP and enables InfiniBand protocols to work over the wide area. RoCE uses much less CPU power than TCP to move the same amount of data and is far easier to tune. One of the projects is led by Dantong Yu at Brookhaven National Laboratory; another is led by Professor Martin Swany at Indiana University; and ESnet staff have participated in the third with Orange Silicon Valley, Bay Microsystems, and System Fabric Works. All three are exploring various ways RoCE can work over a wide area network at 10 Gbps and 40 Gbps. The results have been encouraging. In particular, our work with Orange was the first time RoCE was tested over 40 Gbps and showed that data could be moved at up to 96 percent of the peak capacity of the network. There’s a news release about this on our site.

Q: What’s the plan for the next three years?

Tierney: We plan to upgrade the hosts to Intel’s new Sandy Bridge architecture with the third generation PCI Express bus. Our current testbed hosts have a maximum network I/O rate of around 28 Gbps, but with the new hosts we should be able achieve 100 Gbps with a single host. We are also obviously interested in using the facility to test other new equipment, such as some 40 Gbps NICs (network interface cards) we received from Mellanox. We also plan to test brand new 100 Gbps NICs when they become available from the vendor community.

In the past few years, networking in general and ESnet in particular have gained a much higher profile in the research community. What do you think is behind this and where do you see this leading?

Tierney: When Bill Johnston became the head of ESnet in 2003, he brought a strong research background to the job and began to push the boundaries, looking at what was new and different, and to address the unique needs of the research and education community that were not being met by industry. This led to ESnet taking a more active role in leading novel networking research efforts.

We can also point to Big Data. People are realizing that managing big datasets is a big challenge and that the job of managing the data doesn’t always fall to the site where the data is collected. With more and more collaborations, the data is often not stored where the researcher is.

To efficiently move those large datasets, Eli Dart from ESnet is really pushing the notion that end-to-end network performance means disk-to-disk, not site border to border. This is the “Science DMZ” idea that ESnet developed that is now gaining momentum across the entire research and education community. The NSF’s most recent Campus Cyberinfrastructure–Network Infrastructure and Engineering Program (CC-NIE) is now funding many campuses to deploy Science DMZs. And folks are now looking to ESnet for advice and guidance on how to do this. When it comes to solving end host issues and internal network issues, people are also going to our Fasterdata site, a knowledge database which our team developed and which we know has helped many collaborations to improve their application performance. ESnet has also contributed to the development of tools like perfSONAR for finding and fixing bottlenecks in end-to-end networks.

Lastly, I’ve heard ESnet also has a dark fiber testbed. Can you talk about that?

Tierney: That’s a big topic. Let’s save it for another conversation.

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!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

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

March 18, 2024

Nvidia's latest and fastest GPU, codenamed 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…

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…

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…

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…

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…

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…

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

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…

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…

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…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire