Data Centers Featured Article

Group communications issues are common in cloud computing – and addressing them is key to scaling out the data center.

 

To improve communications, Voltaire Ltd. has introduced a new software and hardware solution that accelerates distributed application group communications in scale-out fabrics.

 

Voltaire is a top provider of scale-out computing fabrics for data centers, high performance computing and cloud environments. Voltaire’s family of server and storage fabric switches and advanced management software improve performance of mission-critical applications, increase efficiency and bring down costs through infrastructure consolidation and lower power consumption.

 

 

The first edition of Voltaire FCA accelerates high performance computing applications like reservoir modeling, fluid dynamics, crash analysis and others by up to a factor of ten. Voltaire’s patent pending Fabric Collective Accelerator (FCA) accelerates MPI collective operations by making use of Voltaire switches and their on-board processors to offload important parts of group communication onto the switching fabric while Voltaire Unified Fabric Manager (UFM) organizes an efficient, topology-based collective flow. Working together, these products make sure that all bottlenecks are eliminated, at the server as well as at interconnect levels. The computational acceleration is accomplished transparently without making changes to the application.

 

Niclas Andersson, technical director at the National Supercomputer Center in Sweden stated that with Voltaire FCA, the IMB Allreduce & Barrier benchmarks worked seven times faster on a 512 core portion of their Intel (News - Alert) Xeon E5520 based cluster. Anderson also added that they were extremely proud to be the first customer to run Voltaire FCA in a large environment, they expect it to enable them to increase the efficiency and productivity of their research clusters by larger orders.

Earl Joseph, program vice president, IDC (News - Alert) had commented that with their global synchronous nature, collective operations often become the obstacle while scaling MPI-based parallel applications to thousands of computers, and have been known to have a considerable impact on the scalability of some applications with standards-based clusters. Joseph added that Voltaire’s FCA provides an innovative way to help in eliminating the congestion created by collective operations in HPC and is also applicable to other distributed computing frameworks like MapReduce in cloud computing.

 

Until recently, typical solutions for accelerating collectives over standards-based clusters were based on host-based offload, solving only a fraction of the challenge at the server level. The Voltaire fabric-wide solution optimizes Voltaire’s unique switch design, which includes a CPU on every module, removing congestion and reducing discontinuation throughout the fabric. Voltaire FCA works with Voltaire’s complete portfolio of 40 Gb/s InfiniBand switches and will be available by early Q2 2010.