[one-users] Open Nebula and Scheduling

[Brian Smith]

Hi, All,

I'm researching the possibility of using Open Nebula as a tool to build
a cloud for HPC purposes.  What I'd like to do is use existing HPC
infrastructure to run virtual machine instances and VM-based clusters
for specific applications, the goal being to reduce lead time for
application deployment and to support applications that have very
specific platform requirements that don't necessarily mesh well with our
existing HPC setup.  

What I'd like to do is continue to schedule jobs against the existing
HPC environment using a traditional batch scheduler like Torque/Maui and
use something like Open Nebula to request, from the batch scheduler,
advance reservations for resources to run VMs.  Obviously, the needed
infrastructure is already in place (multi-core cluster nodes with gobs
of memory, network, etc.).  

My main interest is in whether or not either 

a) Open Nebula can interface with a third-party scheduler to schedule
the standing-up of VM instances or

b) It would be possible to extend open nebula to interface with a
third-party scheduler and which bits of the code base I should be
interested in for this.

The goal, from the end-user's perspective, is to be able to continue to
schedule and run traditional bare-metal HPC jobs through Torque/Maui
while at the same time, new users who have specific application needs
can say "I need X cpus, Y RAM, and Z nodes to run my application for the
next month" and have Torque/Maui handle the reservation or the resources
while Open Nebula takes care of the rest.  

Any pointers or advice would be greatly appreciated!

Best Regards,

Brian Smith
University of South Florida