Parallel Virtual Machine (PVM) for x86 and a C based dialect I've forgotten the name of on a MasPar system. Now get off my lawn, kids! ๐ด
Parallel Virtual Machine (PVM) for x86 and a C based dialect I've forgotten the name of on a MasPar system. Now get off my lawn, kids! ๐ด
As I continue switching my mindset to objects, this space looks even more appealing and that's before considering the huge cost benefits compared to purchasing large parallel filesystem appliances.
Stand up dedicated NFS storage nodes for groups and/or projects, allowing them to run jobs on the storage nodes so they can manage data staging/curating directly from the storage node via the scheduler and have no noisy neighbors to compete with when using it so I/O becomes more predictable.
Nodes share /local/scratch via NFS, allows a multi-node job to use one or more of it's nodes as dedicated NFS servers for the rest of the nodes in the job to use.
Just doing the needful to show users how to simply stage results/checkpoints from node local storage back to a shared filesystem space as part of the job (trap is handy for this in a preempted environments)
Just doing the needful of showing users how to stage input data onto local scratch using simple rsync based cache update approaches.
Those of us who have spent a lot of time supporting the "long tail" on really tight budgets, this post is a little bit of validation. I've had luck in the past with lots of workloads using several different approaches to this, either closely or loosely related:
ClusterShell 1.9.3 is now available in EPEL and Debian. Not using clustershell groups on your #HPC cluster yet?! Check out the new bash completion feature! Demo recorded on Sherlock at @stanford-rc.bsky.social with ~1,900 compute nodes and many group sources!
asciinema.org/a/699526
The upside and downside to democracy is that people get the government they deserve.
Starter kit was a big help!
It's good to be here. I may actually post here instead of just existing to hold onto my name like I did on Twitter/X.