I'm trying to setup a nfs share for a compute grid where I'm expecting to have about 20-30 or 240-360 cores. I currently have one drive (each server has 4 drives) dedicated to moosefs and mount that to each of the servers. It works pretty well. Small file performance sucks though. Above all it's easy to configure and comes with a nice web interface.
I'm now looking to setup a scratch area backed by 8 SSD drives. Anyone know something else I should look at at? Above all it should be simple to setup and operate! I'm ok if the performance is not the very best! Also last time I used nfs I would occasionally get stale file descriptors and I would have to reboot to get rid of them. I've never had that problem with moosefs. So my question is
1. Can NFS handle 240-340 grid jobs doing concurrent reads and writes. Does it buffer request to make sure there is no contention for resources.
2. If so, what NFS solutions should I be looking at?
I'm now looking to setup a scratch area backed by 8 SSD drives. Anyone know something else I should look at at? Above all it should be simple to setup and operate! I'm ok if the performance is not the very best! Also last time I used nfs I would occasionally get stale file descriptors and I would have to reboot to get rid of them. I've never had that problem with moosefs. So my question is
1. Can NFS handle 240-340 grid jobs doing concurrent reads and writes. Does it buffer request to make sure there is no contention for resources.
2. If so, what NFS solutions should I be looking at?