VMware: InfiniBand and RDMA Better Than Ethernet
Last month, VMware’s Josh Simon’s attended the OpenFabrics Alliance User and Developer Workshop in Monterey, CA. While at the event, Josh sat down with InsideHPC to discuss VMware’s current play in the HPC space, big data and the company’s interest in InfiniBand and RDMA. Josh believes that by adopting RDMA, VMware can better manage low-latency issues in the enterprise.
Josh noted that RDMA over an InfiniBand device aids in virtualization. VMware is seeing live-migration times shrinking in its virtualization platforms, such as VMotion and others. The company is also seeing CPU savings in addition, thanks to the efficient RDMA applications it has deployed.
Greg Ferro also posted on The Ethereal Mind blog about how VMware believes InfiniBand over Ethernet is better than Ethernet alone. According to Greg:
“Good InfiniBand networks have latency measured in hundreds of nanoseconds and much lower impact on system CPU because InfiniBand uses RDMA to transfer data. RDMA (Remote Direct Memory Access) means that data is transferred from memory location to memory location thus removing the encapsulation overhead of Ethernet and IP (that’s as short as I can make that description).”
To read the full text of Greg Ferro’s blog post, click here.
To watch Josh’s interview with InsideHPC, or to check out the other presentations from the OFA workshop, head over to the InsideHPC workshop page. Presentations are available for download on the OFA website.
Plugfest 20 Integrators’ List
Last month, the IBTA released the latest update to the Integrator’s List, highlighting all of the products that have been tested and accepted by the IBTA as being compliant with the InfiniBand architecture specification. The latest list contains 353 DDR (Double Data Rate-20 Gb/s)-certified cables, 279 QDR (Quad Data Rate-40 Gb/s)-certified cables, four QDR HCAs (Host Channel Adapters), two QDR switches, one InfiniBand WAN (Wide Area Network) range extender, one SRP (SCSI RDMA Protocol) storage target and one InfiniBand to Ethernet gateway device.
Included in the updated list are InfiniBand Device Vendors:
- Intel (formerly QLogic
- Obsidian Research
InfiniBand Cable Vendors featured on the updated Integrator’s List include:
- Advanced Photonics
- Applied Micro
- Cinch Connectors
- Foxconn Electronics
- Fujikura America Inc.
- Hitachi Cable
- Mellanox Technologies
- TE Connectivity
The IBTA Integrators’ List is designed to support data center managers, CIOs and other IT professionals with their planned deployment of InfiniBand solutions. End users and OEMs reference the IBTA Integrators’ List prior to the deployment of InfiniBand-related systems, including both small clusters and large-scale clusters of 1,000 nodes or more. The Integrators’ List was started in 2003 and currently contains hundreds of InfiniBand-compliant products.
The full list is available at www.infinibandta.org/integratorslist
Plugfest 20 Beta FDR List
FDR cables and devices were also tested during the IBTA Plugfest 20 held at the end of October 2011.The Beta FDR List, which details the results of each product’s performance against the IBTA FDR specifications as well as all the testing performed, is now available for review.
The list features both switches and HCAs and lists the cable and device qualification process for each technology.
Compliant cables include cables from 3M, Amphenol, Cinch Connectors Inc., FCI, Foxconn, Hitachi Cable Manchester, Mellanox, Meritec, Molex and Volex.
The IBTA FDR Interoperability List also featured cables from companies including 3M, Amphenol, Cinch Connectors Inc., FCI, Finisar, Foxconn, Hitachi Cable Manchester, Mellanox, Meritec, Molex, Samtec, TE Connectivity and Volex.
Click here to view the full Beta FDR List.
This month the 21st IBTA Plugfest took place at University of New Hampshire (UNH) Interoperability Lab (IOL) providing participants with an opportunity to measure their products for compliance with the InfiniBand™ architecture Specification as well as interoperability with other InfiniBand products. Stay tuned for a recap in next month’s newsletter.
Plugfest 22 is scheduled for the week of October 15, 2012. If you have any questions, comments, or any other information related to the Plugfest, please email: firstname.lastname@example.org. Results of all past Plugfests are available on the IBTA Integrators List site under the archive section.
OFA Training Opportunities
The next Writing Application Programs for RDMA Using OFA Software is May 22-23, 2012 at the University of New Hampshire's InterOperability Lab (UNH-IOL).
Date: May 22-23, 2012
Location: University of New Hampshire's InterOperability Lab (UNH-IOL)
Fee: $2,495 per person
If you are an application developer skilled in C programming and familiar with sockets, but with little or no experience programming with OpenFabrics Software, OFA RDMA (InfiniBand, iWARP, and RoCE) training is the perfect opportunity to develop your expertise. Each two-day class provides application developers with classroom instruction and hands-on experience coding applications directly to the Verbs API.
Join instructor Dr Robert Russell from the University of New Hampshire Interoperability Lab to learn about coding applications directly to the Verbs API using the OpenFabrics Enterprise Distribution (OFED™) and related software tools.
Reasons to attend:
- Receive detailed classroom instruction in writing an application to the verbs API
- Receive hands-on experience writing, compiling and executing an application program using the OFA stack and software tools
- Access the OFA cluster at UNH-IOL which includes equipment from all of the major InfiniBand and iWARP companies
- Receive source code for over 20 RDMA demonstration applications that can be used to jump start your development work and also receive over 500 pages of detailed instruction material presented during the course.
View the Programming Course syllabus.
View the OFA Training Program Writing Application Programs for RDMA Using OFA Software from SC11.
For any questions, please contact Rupert Dance at email@example.com.
SC12 Call for Papers – Deadline Extended
Don’t forget that full technical papers are due May 4 to be considered for speaking sessions and presentations at SC12. If your company is interested in submitting a technical paper on one of the nine HPC-related topics, visit the SC12 website for more information. New features in 2012 include:
Papers must be submitted electronically via the website https://submissions.supercomputing.org/
- Eight topical areas for presenting the highest-quality, original research: Algorithms; Applications; Architectures and Networks; Clouds and Grids; Performance, Energy, and Dependability; Programming Systems; Storage, Visualization, and Analytics; System Software
- “State of the Practice” integrated into technical papers as the ninth topical area for presenting contributions that can advance the practice of high performance computing, including facilities, services and systems.
- A rebuttal process to enhance the rigor of peer review to determine acceptance of submissions.
Learn more about the submission process and requirements here.
And don’t forget that this year’s Supercomputing Conference is happening November 16-18, in Salt Lake City, Utah. Learn more about the event here.