Simultaneous I/O degradation

[christian]

Running CentOS 4.4 w/ GPFS 3.2.0..1 managing a fiber-attached array consisting of 4 LUNs. I've tried kernels 2.6.9-42.ELsmp & 2.6.9-67.0.15.ELsmp (compiled from src RPM) Performing a dd write from /dev/zero to the mounted array w/ no other I/O averages out to 377MB/sec. dd from /dev/zero to local disc averages out to about 70MB/sec. When I start a local write followed by a write to the SAN, I see about 16MB/sec to the SAN. Local performance is about the same.

I have also run the same test on a SLES 9 SP4 machine fiber-attached to the same array running a 2.6.5-7.286-smp kernel and saw no noticable degradation. To further test it, I pulled the SLES kernel image and modules over to my CentOS box and re-ran the test. Again, no degradation.

I have been trying to get a 2.6.18 kernel compiled, but I can't insert the GPFS modules w/o getting a semaphor mismatch error. Anyone know of any issues w/ the 2.6.9 kernel that might cause this? Any suggestions?