Can’t view this HTML email? Click here.

InfiniBand Trade Association E-Newsletter
img

CONTENTS

IBTA Updates
RoCE Updates
InfiniBand Updates
Member Announcements
Upcoming Events
 
BLOG HIGHLIGHT

Check out the latest happenings within the IBTA and InfiniBand community by viewing our blog.
img

September 2017

IBTA Updates

SDC 2017 Ethernet Storage Fabrics Presentation Now Available

Storage Developer Conference (SDC) 2017, the destination for technical discussions and education on the latest storage technologies and standards, was held September 11-14 in Santa Clara, CA. Representatives from the IBTA led a session on Ethernet Storage Fabrics: Using RDMA with Fast NVMe-oF Storage to Reduce latency and Improve Efficiency.

The presentation provided the following:

  • An overview of NVMe-oF, a new memory standard that allows the latest storage class memories to be extended and shared
  • Review of the architecture of PCI-Express based Flash and SCM storage and how the hardware/software interface is optimized for parallelization and minimization of context switches and interrupts
  • Explication of how this interface can be extended over an Ethernet fabric using RDMA and explanation of the technical and business advantages of this approach over using legacy SCSI based SAN technologies such as Fibre Channel
  • Review of the network requirements to build a true Ethernet Storage Fabric.

Download the complete slide deck from the RoCE Initiative Resources page.


InfiniBand Updates

Biowulf Supercomputer at NIH Gets an Upgrade

CRSA has updated the Biowulf supercomputer with a second increment. The InfiniBand-based cluster is located at the National Institutes of Health (NIH) Center for Information Technology, and was created to process a large number of simultaneous computations in biomedical research areas. This second increment, which will enable researchers to make advancements in new medical treatments, includes:

  • 1,104 compute nodes (1.2 Petaflops) and 72 GPU nodes added to the existing 2,372 node cluster (1.6 Petaflop)
  • An additional 4.8 Petabytes of storage

Read insideHPC’s article for more information on the recent upgrades to the Biowulf supercomputer.

IBM’s New AI Platform Connected via InfiniBand

In a recent article, HPCWire explores the new IBM PowerAI software platform aimed at producing practical scaling of Artificial Intelligence (AI) models for today’s fastest GPUs. PowerAI distributed deep learning (DDL) is a newly developed communication algorithm housed in a communication library. This new technology takes the best of traditional high performance computing (HPC) and combines it with AI to deliver deep learning solutions. To test the performance of this new library, IBM used InfiniBand-based clusters, resulting in a remarkable 95 percent efficiency.

For a detailed overview of IBM’s PowerAI software platform, check out HPCWire’s coverage.

InfiniBand Makes Its Way into Orbit

An InfiniBand-connected cluster has become the first HPC system into space. HPE and NASA have launched the general purpose Spacebourne Computer into orbit aboard a SpaceX Dragon capsule. It will determine if systems such as this one are prepared to effectively operate in the rugged environment of outer space. The spacecraft will circle the Earth’s orbit for a year and run three HPC relevant and realistic NASA benchmarks. Ultimately, HPE and NASA hope to use these test and system design for Mars exploration in the not so distant future.

For more on NASA’s InfiniBand-based space supercomputer, read HPCWire’s article.

A Path to the Cloud for HPC

As more enterprises look to leverage high performance capabilities to process data for Big Data and IoT analytics, there will be a push from cloud providers to offer HPC infrastructure as a service. One example of this trend can be seen in Europe, where organizations will soon have access to HPC infrastructure through an InfiniBand-based 100 Gbps EDR Open Telekom Cloud service, providing high performance local storage through a parallel file system. Using the cloud lowers cost of infrastructure needed for standard workloads, and could do the same for HPC.

Learn more about the emergence of HPC cloud services in IT Business Edge’s article.


RoCE Updates

11 Myths about RoCE

Remote Direct Memory Access (RDMA) first became widely adopted in the HPC industry with InfiniBand, but is now being leveraged by cloud, storage, and enterprise Ethernet networks with RDMA over Converged Ethernet (RoCE). Although RoCE has been well-received by the enterprise storage and networking industry, especially by those wanting to accelerate application performance without overhauling their existing Ethernet infrastructure, there’s still some misinformation that remains about the technology. IBTA Marketing Working Group Co-chairs Bill Lee and Robert Lusinsky took to Electronic Design to dispel 11 of the most common myths about RoCE out there today.

Read the full article on Electronic Design or download the PDF on the RoCE Initiative Resource page.


Member Announcements

Mellanox Adapters for RoCE Ethernet Cloud Power Alibaba Data

Alibaba Group, the world’s largest online and mobile ecommerce company, selected Mellanox 25GbE and 100GbE ConnectX®-4 EN family of Ethernet adapters to accelerate in their Ethernet-based data centers. Alibaba’s decision to deploy RoCE will improve tier overall Ethernet cloud infrastructure, delivering increased system efficiency and enabling application workloads to be highly scalable with high bandwidth networks.

Read Mellanox’s full announcement for more details.


Upcoming Events

IBTA Plugfest 32

October 16 – 27, 2017

The 32nd IBTA InfiniBand and RoCE Plugfest, taking place from October 16 – October 27 at the University of New Hampshire (UNH) Interoperability Lab (IOL), is now open for registration. This event provides an opportunity for participants to measure their products for compliance with the IBTA specifications and includes both InfiniBand and RoCE interoperability testing.

More information on registration and Plugfest 32 can be found on the IBTA Plugfest page.

Questions, comments or any other information related to the Plugfest, please contact ibta_plugfest@soft-forge.com.

Supercomputing Conference 2017

November 12-17, 2017

The 30th iteration of the International Conference for high performance computing, networking, storage and analysis will take place in Denver, CO. SC17 will feature a special technical program, exceptional industry and research exhibits, comprehensive education programs, and many networking opportunities.

If you are planning to attend SC17 and would like to engage with the IBTA, please contact ibta@nereus-worlwide.com.

More information on SC17 can be found on the event website.

For questions or comments, please contact info@infinibandta.org.

img img
img