Hi Robert, This brought a question to my mind - does pvfs tend to use a lot of small files (suggesting to go with reiser 3) or fairly large files on average (xfs) ? I imagine the choice of file system could impact performance.
Cheers, Alan. On 12/5/06, Robert Latham <[EMAIL PROTECTED]> wrote:
On Tue, Dec 05, 2006 at 09:52:43AM -0600, Cliff Kirby wrote: > I have the PVFS server software and Kernel modules compiled and loaded on > the four servers and each one sees the 2 Xraid LUNs as /dev/sdb and > /dev/sdc. Can I setup the PVFS I/O servers to perform any I/O operations > directly to the devices over the SAN and not the GigE network? > Do I have to create a PVFS file system on the SAN device or can I use native > Linux partitioning? Do note that there's no file system driver for OSX clients yet. You will have better luck if you want to use MPI-IO and the userspace pvfs utlities. PVFS will definitely play well with the SAN, but you might have to re-think your toplogy a bit. Start with the SAN file system. Format it however you like. There's no 'mkfs.pvfs' for example: PVFS sits on top of an existing linux file system. You'll need linux servers to attach to the SAN. It's those linux servers your clients will talk to. I don't know of too many people using Xsans and PVFS: I fully expect it to work, but we may not be able to offer much in the way of tuning or configuration hints. However, feel free to carry out a discussion of those aspects on this list, as the resulting conversation should prove quite helpful for anybody following in your footsteps. Have fun ==rob -- Rob Latham Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF Argonne National Lab, IL USA B29D F333 664A 4280 315B _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
_______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
