First off, I have a fix for this. Basically, you need to add the Trond/Higgin NFS patches to the Kernel SRPM and recompile. I'm happy to provide you with my `spec' and `patch' file if you want them. Just send me e-mail. Please try to put this patch into an up-coming `errata' kernel release. We had more problems with certain NFS corruptions after upgrading to the current errata kernel. The situation is this: I have a program that generates large files [100Meg] and performs `fseek's to re-write data somewhere back in the file. Then there's a read program which verifies the data. Both programs are available on request. Here's a table of the results on various NFS client/server combinations: NFS Client NFS Server Results ======================================================== RedHat 4.2 Solaris 2.6 Works fine RedHat 6.1 Solaris 2.6 Works fine RedHat 6.1 RedHat 6.2* Works Poorly *** RedHat 6.1+ RedHat 6.2+ Works OK (fewer corruptions) RedHat 6.1+ Solaris 2.6 Broken (any "fan in") RedHat 6.2+ Solaris 2.6 Broken (any "fan in") RedHat 6.1+ Solaris 2.6+ Broken (any "fan in") RedHat 6.2+ Solaris 2.6+ Broken (any "fan in") Notes: RedHat 6.1 "Vanilla" RedHat 6.1 RedHat 6.1+ Updated RedHat 6.1 (kernel 2.2.16) RedHat 6.2+ Updated RedHat 6.1 (kernel 2.2.16) Solaris 6.2 "Vanilla" i86 Solaris 6.2 Solaris 6.2+ i86 Solaris 6.2 + patch cluster *** The corruptions occurred about %40 of the time and required 10 clients to be writing to the server at the same time.
Created attachment 5563 [details] NFS v2/v3 patch from Trond/DHiggen
I also updated these packages incase I ever want to try NFS v3 (note that all my tests were with v2). The funny version numbers are because I had to hack a few package spec files to build under RedHat 6.x (I'm happy to send specs if you want them): losetup-2.10m-6.1 mount-2.10m-6.1 nfs-utils-0.2.1-2mdkfr nfs-utils-clients-0.2.1-2mdkfr am-utils-6.0.4s5-8fr