As it turns out, you shouldn't try to share out a PVFS directory. PVFS doesn't play well with NFS in that way. After I wrote you about that, I asked some others about it and discovered that you shouldn't do that. Sorry 'bout that!
Becky -- Becky Ligon PVFS Developer Clemson University 864-650-4065 > Becky- > > Thanks for the suggestion. We won't be doing much i/o on /home; mostly I > just need it to be consistent across nodes on an expanding cluster. > > Could you explain how to share out the subdirectories like you mention > at the end? I can't seem to directly mount a subdirectory like > <path>/mount1/home, only <path>/mount1. > > Thanks, > -crispy > > > On 10/21/2010 03:14 PM, Becky Ligon wrote: >> You can define two mountpoints in your pvfs2tab file, say<path>/mount1 >> and<path>/mount2, each pointing to the same tcp address and filesystem. >> This way you can "share out" the different mount points. However, the >> data in<path>/mount1 and<path>/mount2 is being handled by the one set of >> servers and resides in the same physical space, i.e.,<storage >> path>/<filesystem ID>. I know this works, because I tried it when I was >> investigating the multiple ports/alias for you. >> >> While Kevin is working on a bug fix for you, try changing your fs.conf >> file to have only one filesystem. Re-create your storage space. Setup >> your pvfs2tab file with two entires, one for mount1 and one for mount2. >> See if you can't create files using either mount point. >> >> With this approach, an ls on<path>/mount1 and an ls on<path>/mount2 will >> show the same entries. You might try creating two directories, >> <path>/mount1/home and<path>/mount2/scratch, and "share out" >> <path>/mount1/home as /home and<path>/mount2/scratch as /scratch; >> however, I think PVFS doesn't perform well in this situation. In fact, >> PVFS is not intended for a lot of small-io. You might want to make >> /home, >> a regular ext3 (or whatever) directory and let /scratch be a PVFS >> filesystem to be used only for large outputs. >> >> Becky >> > _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
