HPC, XEN, GridEngine and al...

Time for a little thought game, again: Okay, it´s an old story, that you have an vast amount of compute power under your desk, and it´s surely underutilzed as long you don´t start to play games at work. There was the old idea of using workstations with the Grid Engine but nobody really do it. Nevertheless: A dual core Opteron in your workstation is really powerfull, and it would be a waste to allow them to idle.

Okay, now think about the following concept: You install OpenSolaris or (later on) Solaris with XEN on the users desktop, the user doesn´t get access to this hypervisor. This hypervisor contains a grid engine configured as an execution host.

On top of this you install the favourite desktop operating system into an virtual machine. The user can do everything she/he wants with this domain, as long she or he doesn´t take a DVD and instalall the system from scratch (a BIOS password and the correct boot sequence should do the trick)

Okay. For your HPC needs you create an generic installation containing the Operating Environment, the Grid Engine and the common toolset of your HPC environment. This installation is created by using a ZFS emulated volume on a fileserver. After this you create a large amount of clones to use them to boot a HPC DomU on all this workstation.

Now think about the following use of this concept. Whenever you need additional horsepower for your computational work you use the the grid engines in the hypervisor environments to startup HPC DomU´s in the needed amount. After a while the DomU´s are up and running. Now you use the Grid Engine in the HPC DomU´s to execute your computational tasks.

As you loose some computational power by using virtualisation it´s perhaps not the solution for your primary compute farm, but by using such an environment you can use all the horsepower under the desk to react really fast on surges in the comsumption of processor cycles. Using a virtualized domain ensures, that you work in a tightly controlled environment without the need to control the user workstation installation. Using iSCSI and other networked filesystems ensures, that you don´t store data on the workstation and you doesn´t need to roll out new images after a change.

And additionally by using XEN you can do another trick: Whenever a user press the off key of her or his workstation, you can trigger an XEN live migration to another hosts. So it wouldn´t harm you, when the user switches off his workstation.