[Gluster-devel] Scheduling based on network speed / mixing compute and storage nodes

[Jordan Mendler]

This question is a bit more distant. Is there a way to have the scheduler
find the fastest-link/nearest storage brick and send files there? I ask
because we have a grant pending which would allow us to build a ~128 node
compute-cluster, but storage will also be a very large factor. My thought is
that by having compute-nodes also act as storage nodes we can more
efficiently utilize hardware (of course using a large # of AFR's for
safety), and buy more nodes that will be better utilized rather than
spending an extra ~$200-300/TB on storage servers/enclosures.

So coming back to network-speed scheduling, would there be a way to have
each node prefer writing to it's locally hosted gluster brick, to then be
AFR replicated to its close-by nodes that are on the same switch? Also, has
anyone attempted this kind of combined setup of Gluster across
compute-nodes?

Thanks so much,
Jordan