Craig, > I know that PVFS2 does not support locking and understand the > benefits of this design. However, I have a question about what > happens to file writes for contiguous, non-overlapping writes.
I think Avery has patches to support locking at the file/byte range level. I am not sure if they are in some private branch though and it is likely they may need some additional work to integrate with the VFS...Avery, Rob? > > Lets say I have N clients, each writing M bytes. The offset > for each write is n*m. However, what if m*n is does not > fall on either a page or block boundary? Two processes are > going to want to touch the same page in memory (or on disk). > Although the byte ranges are different, is this behavior > defined? Apologies if I misunderstood your question. As long as multiple processes/nodes write to non-overlapping regions, PVFS2's behavior and contents of the file should be well-defined. It does not matter if the byte offsets etc do not align to a page or block boundary since PVFS2 really does not have/impose a native block boundary artifact. PVFS2 builds its file abstraction on top of local disk file systems that are usually POSIX compliant. Since local disk file systems work as expected if multiple processes write to non-overlapping regions of the file even with the page/block boundary non-alignment, PVFS2 should also work the same way. If the backing file system is a clustered SAN file system that for some reason does not guarantee defined behavior for such a workload, I guess PVFS2 will also suffer.. Hope this answers your question. thanks, Murali > > Thanks, > Craig > > -- > Craig Tierney ([EMAIL PROTECTED]) > _______________________________________________ > 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
