We're seeing an extreme performance problem when writing to an NFS-exported directory (exported from Gateway 2K E-4300 system running RedHat 6.1 plus most every issued rpm package to date) with any Sun Solaris workstation running Solaris 2.5.1 or 2.6 mounting that directory. The apparent behavior is that it takes minutes to copy 1-5MB of data to the directory but only seconds to copy the same data from the directory. The RedHat 6.1 Linux system is running on a 450MHz Pentium III with a 12GB disk drive that is nowhere near full across a 10Mb Ethernet, small, private network with near zero contention. When using "top" we see next to zero CPU activity during all of this. We observe, empirically, that data flow appears to be fast but the ending status indication is inexplicably delayed. This may be related but we also get the "program not registered" error message when running "rpcinfo" against the Linux NFS server from anywhere else within that private network including the Linus NFS server itself.
A patch has been available from Sun to fix NFS problems for over a year now. It is not a Linux bug