I think there are two separate areas which this question might be dealing with.
1) Moving nodes/blades back and forth between HPC (we run our HPC as a parallel computing, distributed memory system batch computing with LSF as the queue manager) and VCL. Since we have racks which can do either, we actually reassign nodes/blades to one pool or the other depending on demand - and on the suitability of the hardware - we tend to keep the lower performance hardware in the VCL. I think we actually move a chassis at a time. We don't do this for short time changes (e.g. weekends) - so it is primarily the summer - when the VCL demand is down because of fewer courses and the graduate students and faculty demand is up - because of fewer courses. :-) I think we also used to do this for Winter break - but I don't know if we do that now - the problem being that HPC jobs are running longer and longer, and Winter break isn't long enough to accommodate many of these long jobs. Still, this reassignment significantly adds to the % of the time that the equipment is utilized and so gets more production out of our investment. What is involved is primarily reconfiguring the communication switches so that either: a) the VCL and its management nodes "see" the chassis and then they start assigning reservations to it - or in the other direction b) the LSF "sees" the chassis and assigns it to its queues This is done by remotely reconfiguring the switches, no physical wiring changes need to be done. 2) Working *within* the VCL to assign a cluster. This could be used for HPC production computation, but most often it's used for experimenting with HPC. Sometimes people want to tinker with the OS - and in our HPC area they don't get to do that - they get our vanilla Linux OS. Sometimes, e.g. when developing code, they want to interact with nodes in their cluster to find out what's going on, and they don't get to do that in our *batch* HPC area. There are many other "one off" times when this can be useful - and the VCL allows making a Reservation for a "cluster" if the person has the correct privileges. Aaron Peeler has told me that creating this type of VCL cluster is also called adding subimages and that there is documentation at: https://cwiki.apache.org/VCL/creating-a-cluster-enviroment.html I'm not sure what UTEP is doing with their VCL in this cluster area - perhaps someone from there can fill us in. Also, if I got anything above wrong or omitted something relevant/important - please fill us in. --henry schaffer On Fri, May 17, 2013 at 3:32 PM, Dmitri Chebotarov <[email protected]> wrote: > Hi > Is there a manual which describes how make HPC reservations with VCL ? > I see that code/database mentions 'hpc' and 'vclhpc' projects and would > like to get more info on the subj if possible. > > -- > Thank you, > > Dmitri Chebotarov > VCL Sys Eng, Engineering & Architectural Support, TSD - Ent Servers & > Messaging > 223 Aquia Building, Ffx, MSN: 1B5 > Phone: (703) 993-6175 | Fax: (703) 993-3404 > > > >
