The Linux Beacon--Red Hat and Platform Computing Partner for Supercomputing

Newsletters	Subscriptions	Forums	Safari	Store	Career	Media Kit	About Us	Contact	Search	Home

Volume 4, Number 43 -- November 27, 2007

Red Hat and Platform Computing Partner for Supercomputing

Published: November 27, 2007

by Timothy Prickett Morgan

Platform Computing, one of the pioneers of grid computing for supercomputing environments, and Red Hat, a pioneer in commercially supported Linux distributions and the revenue and market share leader for commercial Linux for a decade, have teamed up to bundle and promote their respective products among the high performance computing centers of the world. Red Hat and Platform Computing have been working together since the latter company rolled together a bunch of open source tools into an integrated stack back in July 2006, called the Open Cluster Stack.

Back then, Platform's OCS product was launched on Dell servers and certified to run on Red Hat Enterprise Linux or its CentOS clone on X86 and X64 servers. The stack included development tools and utilities, file systems, system management, workload management, and resource management tools for both the server and cluster layers of supercomputer clusters. The heart of the OCS product is an open source cluster management project called Rocks, which was created by the San Diego Supercomputing Center and owned and administered by the University of California at San Diego. Platform has also created an open source cluster workload management tool called Lava, which is central to the OCS product and which provides some of the functionality of Load Service Facility, Platform's original (and closed source) grid computing product. LSF can plug into the OCS framework, as can popular open source tools such as Ganglia (which monitors workloads on nodes and clusters), CluMon (which is a system admin dashboard for monitoring grids), MatTool (for managing disks and DNS servers), the Nagios server management tool, and about a dozen other snap-ins, including the IBRIX parallel file system, the MySQL datasebase, the Volcano Web portal, and Intel's compilers.

The Platform stack is also going to be enhanced--either directly or indirectly--by the company's recent acquisition of Scali, another pioneer in the grid computing area that has created a sophisticated set of system management tools called Scali Manage for Linux clusters. Scali Manage is a souped-up version of the open source Beowulf clustering that put Linux on the map a decade ago in the supercomputer centers of the world. Platform announced its intent to acquire key assets of Scali at the end of October for an undisclosed amount; the company bought the key software assets behind the Scali Manage product and retained the key development and support personnel as well. Platform says that it expects to integrate Scali Manage with its LSF product by the second quarter of 2008, and it is possible that Scali Manage could be taken open source (but not probable). It seems far more likely that the OCS product will integrate with Scali Manage, much as LSF does. Ethernet, Gigabit Ethernet, InfiniBand, and Myrinet interconnects are supported by OCS.

In any event, the resulting bundle of the two companies' products will be called the Red Hat HPC Solution, integrating OCS with Enterprise Linux 5. Red Hat will be providing 24x7 technical support for the HPC variant, and is presumably kicking some of the money it receives from support contracts to Platform. Red Hat and Platform say that the HPC stack has been certified on a range of hardware platforms and will be available by the end of this year.

Commercial Linux distributor Novell has been a partner of Platform's since 2005, and has a pretty good foothold in the HPC space compared to Red Hat. Both Cray and Silicon Graphics have modified versions of SUSE Linux Enterprise Server 10 on their monster cluster machines, and do not offer Red Hat Enterprise Linux. Platform has chosen Red Hat, and vice versa, to try to take on Novell and its partners in the HPC space. It will be interesting to see if there is a SLES HPC stack in the works. Hard to imagine there isn't, and ditto for Ubuntu and maybe Mandriva Linux as well.

Platform is charging $150 per node per year for supporting OCS, which does not include Linux support. Red Hat has not talked about pricing for the HPC bundle, but considering the price sensitivity and stinginess of the world's supercomputer centers--which generally have nerds who know as much about Linux as the commercial distributors--there is no way that Red Hat can charge as much for RHEL 5 for these customers as it can for general commercial customers. Historically, those supercomputer centers that wanted to pay for commercial support bought workstation licenses and put the software on their server nodes, and then turned off all but the most necessary services to boost the efficiency of the Linux running on the nodes. Red Hat is charging $299 for a standard subscription to RHEL Desktop 5, which offers 12x5 business support; a standard subscription for RHEL 5 on a two-socket server costs $799. Microsoft is charging $469 per node for a license to Windows Compute Cluster Server 2003 license. It is a fair bet that Red Hat HPC Solution will cost $449, which is just the cost of RHEL 5 support on a workstation plus the cost of Platform's support.