IBTA Plugfest #29 April 4-5, 2016 – Register Today!
Next Integrators’ List and RoCE Interoperability List on the Horizon
Registration is now open for the 29th InfiniBand Compliance and Interoperability Plugfest! Join us April 4-15, 2016, at the University of New Hampshire (UNH) Interoperability Lab (IOL) where InfiniBand vendors will gather to measure their products for compliance with the InfiniBand architecture specification. Participants will also have the opportunity to test their products’ interoperability with other InfiniBand and RoCE solutions.
Additional event details and registration information are available on the Plugfest page.
IBTA is also close to finalizing the results from Plugfest 28, so stay tuned for an update on the availability of the latest Integrators’ List and RoCE Interoperability List.
IBTA Wants You – Guide the Future of InfiniBand, RoCE and Performance-Driven Data Centers
For any organization, a new year provides a great opportunity to reflect on the past and set a stronger course for the future. Companies can take a variety of internal actions to prepare for impeding market changes, but rarely do they have the power to influence the course of an entire industry on their own. For those devoted to improving clustered server and data center performance, joining an industry alliance such as the IBTA offers a chance to contribute to the foundational work that sets the path for technological advances one, five and ten years into the future.
For more on the benefits of IBTA membership, check out our blog.
InfiniBand Enables Intelligent Networks
Gilad Shainer, of IBTA member company Mellanox, recently wrote a guest article about the high-performance computing industry’s ever-growing need for faster, more reliable and more efficient systems. He highlighted several paradigm shifts within the industry to address this demand, focusing his attention on the most recent shift to Co-Design architecture, which creates synergies between the software and the hardware. Co-Design treats the network as a co-processor that offloads tasks from the CPU, enabling the CPU to concentrate on processing data and executing applications. This requires an intelligent interconnect, such as InfiniBand, to operate as a co-processor and improve overall performance.
Read the guest feature on insideHPC to learn more.
Need for Speed: Comparing FDR and EDR InfiniBand
A recent two part Dell TechCenter blog series compared the performance between the latest InfiniBand data rates, FDR and EDR, via adapters from Mellanox. In part one, they tested the latency, bandwidth and HPL results using industry-wide micro-level benchmarks and applications on HPC cluster configurations. Part two shares more data comparing the performance of FDR and EDR InfiniBand, this time leveraging additional real-world applications such as ANSYS Fluent, WRF and NAS Parallel Benchmarks.
Check out the detailed results of Dell TechCenter’s evaluation here (Part 1, Part 2).
RDMA Interconnects Paving the Way for NVMe over Fabrics Technology
NVM Express (NVMe) technology is quickly being recognized as an innovative, high performance interface for solid-state drives as well as newer persistent memory solutions. Given the improved capabilities of NVMe, high performance and low latency networking is required. In response, the industry developed an NVMe over Fabrics (NVMe/F) specification to allow for flash devices to communicate over RDMA fabrics, including InfiniBand and RoCE. With NVMe/F support for RDMA, remote NVMe access across a network can experience the same low latency as local attached NVMe.
Read the RoCE Initiative blog post to see why 2016 will be a big year for NVMe/F and RDMA.
Dive into RoCE – Visit the RoCE Initiative’s Resource Library
The RoCE Initiative is committed to increasing RDMA awareness among system architects and IT managers that design, deploy and operate Ethernet-based data centers. A pillar of our educational program is the Resource Library, which contains a growing list of RDMA over Converged Ethernet (RoCE) whitepapers as well as a glossary of relevant industry terms associated with the technology. IBTA member companies generously provided materials to help explain the many benefits and applications of RoCE.
For an overview of our current collection, check out the latest RoCE Initiative blog post.
New Version of RDMA-Spark Runs Native InfiniBand and RoCE
Big Data can quickly overwhelm the latest and greatest networks. The Higher-Performance Big Data (HiBD) team at Ohio State University released RDMA-Apache-Spark 0.9.1, which runs native InfiniBand and RoCE to help alleviate congestion issues and speed up data analytics. According to the HiBD team, the packages are being used to accelerate applications by more than 135 organization throughout 20 countries.
For more on the benefits of the new software, read insideHPC’s coverage.
Checking the Trends at Storage Visions 2016
NVMe over Fabrics (NVMe/F) was the subject of an IBTA-sponsored session from Mellanox’s Rob Davis at Storage Visions Conference 2016. Presented as part of the Emerging Storage Technologies track, the session covered both the NVMe/F protocol as well as the high performance network types that support it, including InfiniBand and RoCE. Rob explained why NVMe/F depends on RDMA transports, calling it the “killer app” that will greatly boost RDMA’s market adoption overtime.
Check out Electronic Design’s event overview for more.
CCRT boosts industrial innovation with a petascale supercomputer from Bull
The French Alternative Energies and Atomic Energy Commission (CEA) and its partners at the Centre for Computing Research and Technology (CCRT) have invested in a new petaflop supercomputer to help support digital simulations for energy, aerospace and environmental industries. Designed by Bull, the COBALT supercomputer will have a peak performance of 1.4 petaflops, making it three times more powerful and more energy efficient than CCRT’s current system. A 100 GB/s InfiniBand EDR interconnect solution forms the core of the new computer, which should satisfy the partners’ needs for the next four to five years.
View Bull’s official announcement to learn more.
Further expansion in the Netherlands of Bull supercomputer in 2016
SURFsara, an organization for ICT in the Netherlands, is set to expand the capacity of its Cartesius national supercomputer in the second half of 2016. Cartesius is primarily used for scientific research by universities and institutions that focus on climate research, water management, improving medical treatments and clean energy. Bull is contracted to update the system using Intel Broadwell processors and InfiniBand EDR, which will increase its speed from 260 teraflops to a total of 1.8 petaflops.
For more on the news, check out the press release from Bull.
OpenFabrics Alliance (OFA) Workshop 2016, April 4-8, 2016.
The OFA Annual Workshop, held in Monterey, CA, is the premier event for collaboration between OpenFabrics Software (OFS) producers and those whose systems and applications depend on the technology. Every year, the workshop generates lively exchanges among Alliance members, developers and users who all share a vested interest in high performance networks. The OFA hosts its workshop annually as a joint effort among open source networking community members. The objective is to create an event driven by the needs and interests of attendees. As a result, the audience profiles continue to expand. Attending software developers and distributers are now joined by a growing number of OFS consumers including:
- Application end users
- Enterprise data center managers
- Network technology researchers
- Networking middleware developers
- Operating system vendors
- Storage Vendors
- System & network administrators
- System OEMs, architects & integrators
For information on the OFA Workshop 2016, visit the event website.
For questions or comments, please contact firstname.lastname@example.org