Red Hat introduces its HPC offering
Red Hat today announced the availability of its High Performance Computing Solution, based around a special edition of Red Hat Enterprise Linux 5.2 (RHEL) and a bundled copy of Platform Computing's Open Cluster Stack 5 (OCS) cluster provisioning and management package. There is also a standalone version of just the OS component, Red Hat Enterprise Linux for HPC Compute Nodes, available in different editions for two-socket and four-socket servers for a minimum of four cluster nodes.
The Compute Node product is a cut-down, but significantly less expensive, edition of RHEL, intended to run on the nodes of a cluster - the dedicated compute engines, typically headless blade or slim rackmount servers managed through a front-end system. Red Hat's bundle starts at $249 per node, with the bare RHEL for HPC Compute Nodes product is $79 per node. Currently, Platform's page on OCS5 lists only certain Dell servers as being compatible and all its illustrations and examples use Dell kit, but OCS5 is part of Intel's Cluster Ready program which lists a variety of other vendors of clustering hardware and software suppliers, so broader hardware support should follow in time. Platform's page on OCS links directly to a Dell page about its a range of ready-made pre-configured clusters.
OCS also works with CentOS Linux, the free binary-compatible distribution created by recompiling Red Hat's publicly-available source code. OCS itself includes of a bundle of HPC-related tools, many of them open source, such as the Nagios server-monitoring program and Platform Lava workload scheduler.
Just a week ago, Microsoft announced its HPC product, Windows HPC Server 2008. Red Hat's rival product works out slightly cheaper including three years of support, but Microsoft says it will clarify its pricing in November.
The products are the latest step in the commoditisation of HPC clusters based on commodity x86-64 computers. Until recently, building a cluster meant building racks of bare-bones PCs, hooking them up with a high-speed interconnect such as Infiniband, Myrinet or Gigabit Ethernet, installing operating systems on all the nodes and assembling the software for a front-end system to control the cluster, assign jobs to the nodes, collect and collate the results and so on. As this is a rapidly-growing market segment and many organisations lack the in-house expertise to do all this, vendors such as Intel, Dell, Platform and Red Hat are working together to assemble complete, ready-to-use cluster products, making supercomputer performance available to smaller customers - and not incidentally, at the same time sell single products consisting of entire racks full of servers and their accompanying software licences.