Berkeley Lab Collaboration with Ciena Enhances Optical Chip Design Process


A unique collaboration between a U.S. telecommunications equipment provider and a Department of Energy (DOE) Office of Science national laboratory has helped dramatically improve design cycle times for future high-speed optical networking components.

wl3-extreme-chipComputer scientists and mathematicians from Lawrence Berkeley National Laboratory’s (Berkeley Lab) Computing Sciences organization worked with engineers at Ciena, a leading networking company, to speed up the process by which Ciena validates the design of its ASIC (application-specific integrated circuit) chips. The collaboration grew out of the existing relationship between Ciena, a pioneer in high-bandwidth optical transport technology, and the DOE’s Energy Sciences Network (ESnet), which uses Ciena products to support its high-speed network.

“Modern science requires fast global networks like ESnet, and almost all DOE network traffic happens to run over optical gear from Ciena,” said ESnet Director Inder Monga. “ESnet doesn’t transport quite as much traffic as, say, Google, but we have the unique challenge of coping with massive single data streams from detectors and instruments like the Large Hadron Collider, the world’s largest and most powerful particle accelerator. This means we need to exploit leading-edge networking components, and we’re often early adopters of those components.”

Read more.

ESnet Begins Testing TCP-BBR


Update to underlying internet protocol promises to improve data transmission performance

With Google’s Sept. 16 release of the new TCP congestion control algorithm, BBR (Bottleneck  Bandwidth and Round Trip Time/RTT), ESnet staff are gearing up to test this promising development.

According to early reports, Google engineers reported that the new algorithm “significantly increased throughput and reduced latency.” Google cited an increase in the speed of data transfers from three megabits per second (Mbps) to 9,150 Mbps on a test link with 1% packet loss and a latency of 100 milliseconds. Early testing by ESnet engineers has shown improvements from 10x to 100x on some international paths, but no improvements on other paths.

“I’m eager to work with our Google collaborators to understand the details of how BBR works and how it may benefit data-intensive science applications in the future,” said Brian Tierney, head of ESnet’s Advanced Network Technologies Group and co-creator of ESnet’s Fasterdata website. “This is potentially a great step forward for TCP, and we’re eagerly testing BBR TCP. We’ll post results as we get them.”

More technical details will be revealed when a description of BBR appears in the September-October 2016 issue of ACM Queue. Among the developers of BBR is Van Jacobson, who was a primary contributor to the original TCP in the 1980s when he worked at Lawrence Berkeley National Lab, where he led the Network Research Group.

Tierney said ESnet has begun testing BBR using the 2,000 instances of the perfSONAR network performance measurement hosts installed on research and public networks around the world. By using BBR to send data from one of ESnet’s test perfSONAR hosts, engineers can get details on the data transfer performance across 2,000 network paths with 2,000 different characteristics. “perfSONAR provides us with a really valuable test environment for new networking innovations like TCP BBR,” Tierney said.

Tierney said the old analogy of the Internet as an information highway really applies to the current version of TCP. As freeway traffic gets more congested, even the slightest event like a car with a flat tire can dramatically compound the situation as everyone slows down to rubberneck or go around.

With the current version of TCP, data traffic slows down when it hits congested stretches of a network and the problem gets worse the farther data travels. On some paths, BBR appears to help mitigate this slow-down.

ESnet is particularly interested in BBR as the network carries massive sets of data between 50 main sites in the U.S. and across four high-speed trans-Atlantic links.

“It looks like anyone who transfers big data over long distances could see huge improvements,” Tierney said. “The longer the path they are covering the bigger the difference will be.”

The announcement about BBR was posted on the Patchwork Linux network development website and can be found at https://patchwork.ozlabs.org/patch/671069/

Watch the ESnet blog for updates as BBR tests are conducted. Results will also be shared on ESnet’s Fasterdata website where the organization houses best practices and a broad knowledgebase focused on high performance network engineering, performance testing and tuning, among other topics.

2016 TechEx Conference Taps ESnet’s Expertise


From determining the future networking needs of researchers to speeding up the flow of scientific data, ESnet staff will share their expertise at the 2016 Technology Exchange being held Sept. 25-28 in Miami. The 2016 edition of TechEx is presented by Internet2 and co-hosted by Florida International University, The University of Florida and Florida LambdaRail.

On Sunday, Sept. 25, Jason Zurawski and Kate Mace of ESnet’s Science Engagement Team led “Demystifying the Science Requirements Review Process for Networking,” a half-day tutorial on the ESnet network requirements review process in the context of the Research and Education (R&E) networking community.

Brian Tierney, head of the Advanced Network Technologies Group, is one of three speakers presenting an overview of the newly released perfSONAR 4.0 network measurement software tool on Tuesday, Sept. 27.

Andy Lake of the Advanced Network Technologies Group will present a “perfSONAR 4.0 Feature Tour,” highlighting many of the important new additions and enhancements found in perfSONAR 4.0. It will cover topics such as improvements to MaDDash, updated test result charts, new central test configuration features, and information about OS upgrades.

Tierney will give another talk on Tuesday on “Improving Performance of 40G/100G Data Transfer Nodes,” which will cover host tuning for 40G/100G hosts, as well as new TCP enhancements that can greatly improve the stability of high-speed flows.