ScaleMP sells a virtualization solution that enables you to create one big virtual server out of several smaller physical servers (previous coverage). It’s sort of the opposite of what most virtualization vendors do, where you run multiple virtual servers on one physical server. The latest version of ScaleMP will now enable users to run other virtual machines from within the big virtual machine.
Unlike a Beowulf cluster, a ScaleMP virtual machine acts as a single server with a single operating system image which takes advantage of all the physical machines’ RAM, CPU and other resources as a single pool. The general reason you’d want to run ScaleMP is that you have high performance computing needs and want to save money on servers. The cost of servers isn’t exactly linear. It’s cheaper to buy four servers with eight sockets each than it is to buy a single eight socket server. You can use ScaleMP to effectively glue all those two socket servers together.
Up til now, you’d have to run everything on those monster servers natively. You couldn’t install a hypervisor and start running sub-VMs. Now you can. This opens a number of possibilities. One use case for ScaleMP is pooling server resources for mega-jobs that don’t need to be done often. For example, maybe you have some big financial data that needs to be crunched once a quarter. You need a supercomputer for this task, but the rest of the year you don’t have such high requirements. You could use ScaleMP to pool a bunch of resources for that quarterly task, and then redeploy them for some other task as necessary later. Using the new VM on VM feature, you can make this re-deployment much easier.
Other new features in the latest ScaleMP release include:
- Parallel I/O support
- Improved MPI offload engine
- InfiniBand-FDR support