QLogic automates installation, verification, monitoring and management of InfiniBand fabrics

17 Jun 2011 Aliso Viejo - Leveraging its unique, system-level understanding of high performance computing (HPC) communications fabrics, QLogic has extended its software capabilities with InfiniBand Fabric Suite (IFS) 7.0. The latest version of QLogic's fabric management software package enables users to optimize fabric performance and communications efficiency, while minimizing management costs, for HPC clusters of any size.

A key, new feature of the release is the integration of vFabric QoS (Quality of Service) with work flow schedulers from Adaptive Computing and Platform Computing into IFS 7.0, making it possible for HPC users to set a priority and QoS level as the message passing interface (MPI) application is being scheduled. This reduces management overhead, simplifies cluster scheduling, and optimizes use of the fabric.

"IDC end-user studies and research have consistently shown that InfiniBand-based systems are the fasting growing in HPC today. Integrating vFabrics with high performing job schedulers is another step forward for QLogic's IFS", stated Steve Conway, Research Vice President at IDC. "This integration joins other key features such as adaptive and dispersive routing to make IFS 7.0 stronger in the InfiniBand fabric management space."

IFS 7.0 incorporates the following new features:

  • Integration with resource management and/or workload scheduling software, the Adaptive Moab scheduler (from Adaptive Computing) and Platform Cluster Manager and LSF scheduler (from Platform Computing) play a critical role in HPC cluster management. Until recently, these cluster job schedulers have been independent from InfiniBand fabric management. Working closely with Adaptive Computing and Platform Computing, QLogic has integrated the QoS capabilities of vFabrics, a core component of InfiniBand Fabric Suite 7, with the job schedulers, to optimize the launching of MPI applications within a QoS priority level. QLogic has extended the integration of cluster and fabric management by allowing QLogic IFS 7.0 Fast Fabric management tools to be directly called from Platform's console management tool.
  • Enhanced Fabric Viewer, an enhanced Fabric Viewer enables users to rapidly identify and correct fabric errors, signal integrity issues and congested links/paths.
  • Broadening industry support: IFS 7.0 supports newer versions of Performance Scaled Messaging (PSM) that is now included with OpenFabric Enterprise Distribution starting with version 1.5.3. PSM is able to take advantage of QLogic's TrueScale IFS pipelined data transfer; enhanced feature which is designed specifically to optimize MPI communications and performance.
  • Fabric Dashboard: part of the Fabric Viewer, the Fabric Dashboard is an intuitive graphical user interface (GUI) that supports statistical data in (near) real time.
  • Fabric deployment: IFS 7.0 increases support for comprehensive stress tests of links and cables to make it faster and easier to bring up InfiniBand fabrics.
  • Improved static routing balance: IFS 7.0 includes a new algorithm that significantly improves static routing balance for 'all-to-all' communications for large fabrics.
  • New Congestion Control Algorithm: the Congestion Control Algorithm delivers a 25 percent throughput improvement over previous IFS releases.
  • Enhanced performance: IFS 7.0 features improvements for enhanced file system performance with Lustre and GPFS as well as industry-leading performance for high core count processors, such as AMD Opteron Magny-Cours and Interlagos.
  • Broad software support: As a proven commercial-grade suite of tools designed to make HPC solutions easier to install and manage, IFS 7.0 supports a comprehensive software ecosystem, including:
  • Continued support for OpenFabrics Enterprise Edition version 1.5.3
  • New support for Red Hat Enterprise Linux versions 6.1 and 5.6
  • Expanded support for Nvidia GPU-accelerated clusters and applications through ongoing support for NVIDIA GPUDirect, and expanded support for CUDA versions 3 and 4.
  • Finally, IFS 7 includes support for the most widely used HPC workload management tools, from Platform Computing's Platform Cluster Manager and the Adaptive HPC Suite from Adaptive Computing, to open source ROCKS Rolls software kits.

This unique combination of features allows HPC customers to obtain maximum performance and efficiency from their cluster investments while simplifying management.

"A continued focus for QLogic is to simplify and improve the management of InfiniBand environments", stated Joe Yaworski, director of HPC product and solution marketing at QLogic. "By automating work flow scheduling and supporting virtually every major ecosystem partner, IFS 7.0 delivers the tools that network managers need to get the most out of their InfiniBand fabric and compute cluster investments."
Source: QLogic