Platform Computing may have rejiggered its grid and workload management software so it can be used to manage clouds of virtualized servers or clusters running latency-sensitive financial services applications, but it has not forgotten its roots in high performance computing. And today Platform is whipping up a more extended set of tools, called Platform HPC Enterprise Edition, to keep supercomputer shops buying its wares instead of looking elsewhere.
Nearly a year ago, when Platform was rolling out its cloudy tools, called Infrastructure Sharing Facility, the company also announced an updated and entry-level HPC management tool called HPC Workgroup Manager, which bundled together Cluster Manager (formerly known as Open Cluster Stack 5), LSF Workgroup Edition, and Platform's own message-passing interface (MPI) parallel computing stack, which it acquired from Scali several years ago. (Subsequent to the announcement of HPC Workgroup Manager, Platform bought HP's MPI stack, which is being borged into its tools.) Workgroup Manager was limited to machines with 32 or fewer server nodes. Starting today, it is rebranded as HPC Workgroup Edition.
That's because there is now something called HPC Enterprise Edition, which has no limitations in terms of the number of nodes it can scale to. The Enterprise Edition includes the full-on versions of LSF (a workload manager for grids), Cluster Manager, and the two MPI stacks, and also chucks in Infrastructure Sharing Facility and the Adaptive Cluster feature that was announced last November to schedule the deployment of bare-metal Windows HPC Server 2008 or Linux instances on clusters of x64 machines. The trick is that Adaptive Cluster is hooked into LSF, knows which applications need what, how many OS images are needed and - based on the workloads in the queue - it can figure out how to get all of the nodes for a job configured just as they are needed. Time to fire all those grad students. . . .
The HPC Enterprise Edition also includes a new feature called Application Center, which is a Web-based graphical user interface for the management suite - for those who are command line-challenged. The Application Center comes with seven popular applications already pre-integrated. These include ones those from HPC biggies such as Fluent, LS-Dyna, MSC Software, Schlumberger Eclipse, Blast, Ansys, and Abacus. The EE version of the HPC tool also includes something called RTM, which is short for Report, Track, and Monitor, and as the name suggests, RTM is a dashboard for cluster admins that allows them to watch workload and system performance and helps them automagically give resources to jobs on the cluster that need more oomph.
HPC Enterprise Edition is supported on x64-based clusters and currently supports Red Hat Enterprise Linux, Novell SUSE Linux Enterprise Server, the CentOS clone of Red Hat, and another clone of Red Hat called Scientific Linux, which was created by CERN and FermiLab. No word on when Microsoft's Windows 2008 HPC Server will be tossed into the mix. Pricing information for HPC Enterprise Edition was not available at press time, but William Lu, director of HPC marketing at Platform said that it will cost more than the Workgroup Edition.
Supercomputer maker Cray, by the way, is selling an OEM version of HPC Enterprise Edition on its new CX1000 supers, which were announced in late March supporting Xeon 5600 and 7500 processors from Intel, and Nvidia Tesla graphics co-processors. Dell will probably cook up a variant for its PowerEdge clusters at some point soon. ®