Platform Computing has released its latest cluster management package, the HPC Enterprise Edition. The new offering is aimed at commercial HPC users who are looking for an easy-to use, integrated software stack for cluster systems large and small.
There are a lot of cluster management packs out there already, but Platform has the unique advantage of being able to offer all the major components of a standard cluster suite. That includes a cluster manager (Platform Cluster Manager), a workload manager (Platform LSF), and an MPI library (Platform MPI, formerly HP MPI). The Enterprise Edition also comes with extra goodies like a workload and system monitoring tool (Platform RTM) and a multi-boot OS manager (Platform ISF Adaptive Cluster) — useful when you want to mix various Linux flavors (or even Microsoft Windows) in your cluster.
A web portal runs on top of the tools and provides a user-friendly interface that ties all the pieces together. Although one cluster does not a cloud make, the web interface and the ability to dynamically switch the OS on individual nodes, based upon the workload, makes the offering at least partly cloudy. “So we can say that this really makes the cluster cloud-ready,” suggests William Lu, director of HPC product marketing at Platform.
Of course, Platform sells the management tools and the MPI library separately, but the idea behind the Enterprise Edition is to bundle the pieces into a turn-key solution for the non-HPC expert. For organizations who are used to rolling their own cluster stacks and executing jobs with scripts, this is probably not the product for them. The Enterprise Edition is really designed for those with minimal expertise in high performance computing and Linux. The web portal, in particular, replaces the need for scripting and allows average users to run and monitor jobs remotely from their deskside machines.
The portal also has another useful feature: application templates. This provides the capability to launch application-specific jobs via a web interface — from across the room or across the country. Platform has pre-integrated support for a number of well-known vendors and applications (Abaqus, Ansys, Blast, Eclipse, Fluent, LS-Dyna, and Nastran), but templates can be set up for any software application. With this facility, the end user can launch and control execution by specifying input parameters, CPU allocations, node allocations, and so on. The user input is transformed into a script underneath the covers, which then launches the app. The portal also lets you monitor job status as well as examine input and output data. That data can be transferred from the cluster back to your local workstation, if need be.
There’s even web portal functionality for the system admin. Using the Platform RTM (Report, Track And Monitor) tool, the administrator is able to keep tabs on applications, as well system parameters like CPU usage, free memory, and I/O status. Floating license usage can also be tracked via this same interface.
One of the more significant aspects to the HPC Enterprise Edition from the customer perspective is its tight integration. Rather than stitching together multiple components from different vendors and open source providers, you have just one product to purchase and install. All the pieces talk to each other in their native Platformese underneath the web interface. From the user’s and admin’s point of view, it’s designed to make the cluster feel like a unified system.
The other nice side effect of an integrated toolset is that if there is a problem (and eventually there always is), the offending software is much easier to identify since there’s only a single vendor to call. In a mixed-component HPC environment with multiple software providers, finding the wayward tool can be a non-trivial task that can involve some vendor-to-vendor finger-pointing. This may seem like a rather mundane concern, but reducing troubleshooting time and operational costs is a big deal for HPC cluster environments.
The HPC Workgroup Edition is a sister product to the Enterprise Edition within Platform’s HPC Suite taxonomy. As reflected in its name, the Workgroup product is targeted to small clusters — 32 nodes and below — and is aimed principally at the research and education community or for users running a single application. This offering is missing the application template feature and the RTM tool, but still retains a web interface. A typical user would be a professor who got some funding for a project and wants to buy his own cluster to develop one or two home-grown research apps. Once you start running multiple ISV codes, or have a larger cluster to manage, it’s probably time to move up to the Enterprise version.
Although you can install the Enterprise Edition on your own stripped down cluster (just the hardware and an OS), most of these will probably be pre-bundled into OEM offerings from Platform’s partners. Cray is already selling them with their CX1 and CX1000 systems. Given that IBM, HP and Dell are also Platform partners, it wouldn’t be surprising to see the Enterprise Edition show up on those vendors’ machines as well.