Platform, Dell push open source cluster management

Platform Computing, one of the pioneers in parallel supercomputer management, this week rolled out the next edition of its open source parallel cluster management tools, Open Cluster Stack 5.

The launch comes just a few weeks after commercial Linux distributor Red Hat stole a little thunder by announcing its own bundle of RHEL 5 and OCS 5 that it hopes will give it more than a toehold in the HPC space.

Dell and Platform have been partners for many years, and as HPC became a strategic area for Dell in 2005, the company hooked up with Platform to bundle its open source Rocks management tools, which are a derivative of technology that was created by the San Diego Supercomputing Center and owned and administered by the University of California at San Diego. With OCS 5, Platform switched the heart of the OCS stack to a new set of code created through the open source Kusu project, which started from scratch to make the management of Linux-based parallel clusters easier to manage. (Kusu itself supports Red Hat, SUSE, and Ubuntu Linuxes.)

Among other things, the OCS 5 stack (all of it open source) has a cluster workload management tool called Lava, which was created by Platform and which is on its proprietary LSF grid management tool; it also uses the Nagios system monitoring tool, the Cacti node and cluster monitoring tool, and the Ganglia workload monitor for the cluster and its nodes.

With the OCS software, Platform has had to pick its partners in an effort to make money. Dell, which doesn't have its own systems software development beyond its OpenManage tools for its PowerEdge servers, and Red Hat, which has a much lower market share in HPC than in the commercial Linux space at large, are natural partners for Platform when it comes to x64-based supercomputer clusters.

IBM, Hewlett-Packard, Silicon Graphics all have their own tools (some of them coming through partners, some through open source tools), and even though they are all partners with Platform, peddling an OCS-Red Hat-PowerEdge bundle gives all three a preconfigured, integrated cluster option with which to pitch to government supercomputing labs, academic institutions, and corporate buyers that is as easy to order and install as the so-called "bright cluster" preconfigured machines being sold by IBM, HP, SGI, and other super sellers.

The OCS 5 stack is available for free if you want to support yourself, and the odds are that plenty of academic supercomputer labs will do just that because they have plenty of grad students they can punish. Corporations would rather pay for software than people, and so would governments. And that is where a commercially supported version of the OCS 5 stack is important. Right now, there are only two ways to get that.

One is to buy the Red Hat HPC Solution, which was announced two weeks ago and which includes the RHEL 5/OCS 5 combo for $249 per server node. The other is to buy OCS Dell Edition, which is tweaked with some Dell-specific code and sold on preconfigured Dell clusters.

According to James Pang, vice president of product management at Platform, the Dell Edition of OCS includes integration with Dell's OpenManage tools, device drivers specific to all the current PowerEdge servers and their related storage arrays, and a Web console for managing the whole thing.

None of this extra sauce is part of the Kusu project and it is certainly not open source. It is, in fact, how Dell expects to get a little bit of leverage in supercomputing labs. Platform does not sell the product directly itself, ironically, and is relying on Dell to do all the selling except for a few whitebox server makers that Platform has partnerships with in Japan.

The list price for OCS support is $150 per server node per year for support on the Dell boxes, which can be Xeon or Opteron machines. A single-node in a Dell bright cluster includes a PowerEdge 1950 III or SC1435 server with two sockets using high-bin quad-core processors, 16 GB of memory, three years of hardware support, a three-year support contract for RHEL 5, and a three-year contract for OCS Dell Edition, all for $4,600. Dell is also a reseller of the LSF tools, if customers decide they want to get more sophisticated workload management for their clusters, and is able to sell an LSF and OCS bundle at a discounted price.

According to Pang, Dell and Platform have been able to install about 100 clusters per year since they established their partnership, and the OCS tools are on a "pretty good percentage" of the bright clusters that Dell sells. Pang says the two companies expect the OCS 5 release to help ramp up sales even further. Platform is also pursuing SUSE Linux customers, and will roll out support for OCS 5 on SLES 10 SP2 in the next few months.

For now, the majority of the OCS-PowerEdge clusters have gone into government and academic research facilities, but Dell and Platform are gaining some traction in industrial manufacturing, oil and gas exploration, and life sciences. ®