Hi David, > I would like to know if anyone on this list has some basic metrics or > feelings about the following:
Unfortunately, I am not sure if I can answer any of your questions specifically, but here goes.. > 1. Performance with many small files (50,000 files that are about 10k > each)? find, ls, things like that. How does it perform? We do have a client-side attribute cache and name cache that should help with reducing load on the server for metadata intensive operations like find or even ls. Hopefully that should help us perform as well as nfs for read-only metadata operations. Since we do not write-back attributes in a delayed fashion, this may not help with update metadata operations. Small file I/O has some special optimizations and tunable knobs (like disabling striping) that should help us do a better job than what we used to earlier, but without adding a client-side data cache with delayed writebacks, performance cannot compete with nfs. I dont have any numbers to substantiate any of the above claims though :( > 2. Can pvfs2 grow? I see you can add new servers to the pool (though > with some downtime/difficulty). Will a more scalable solution be worked-on? Hopefully at some point, we should be able to do this without having to umount the volumes and/or restart clients and servers. Not sure at this point how long it is going to take though... > 3. What's the largest (in terms of nodes and TB) known pvfs2 installation? Hopefully RobL, Pete and/or others on the list can chime in on this. Do share your experiences and/or feel free to ask questions if you encounter any difficulties. Thanks! Murali _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
