On Wed, Dec 20, 2006 at 03:34:43PM -0500, Joe Georger wrote: > From what I understand Lustre offers redundancy in the form of > server failover, and I think I read at their site files can be > striped across luns in a raid5 fashion to give even a little more > protection. I don't think pvfs offers any redundancy yet, but I am > not sure since it's been over 6 months since I looked at it and I am > a newb to boot.
The "raid5 like" striping approach to redundancy has been on the Lustre roadmap for some time. I don't believe they have implemented such a technique yet, but I could be wrong if it's something they did very recently. The server failover options available to Lustre are the same ones available to PVFS. We've been getting dinged on this a lot lately, mostly because we do not have the process well documented or well publicized. Since PVFS clients are stateless, it's easy for us to failover a server to another node. Clients retry their requests, and otherwise carry on as usual. The tricky part is getting the apropriate hardware with shared access to storage. Setting up the linux-ha package to do active-active failover isn't too hard, but we could definitely document the process much better. Don't feel bad about being new to PVFS: if you are evaluating alternatives to Ibrix, feel free to ask the list any questions you've got. ==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
