I've just put my first PO up on OES2 Linux (EXT3 FS on an LVM partition - separate disk from the root partition) in a VSphere VM housed on an ISCSI SAN. Since moving the users over (about 100 connections typical) I've noticed a bit of odd behavior. The IOPS measured on the SAN aren't very high, averaging about 80-100 with a steady 3MB/s transfer rate, but the I/O size is a bit high - about 50KB, compared to <10KB on a Netware PO in the same configuration. It's driving up the SAN latency from <3ms to 8-10ms. I've gone through everything tuning-related I can think of, checked the file system block sizes, enabled noatime on the mount (which helped a bit), all of that. It seems to be related to the bursts of a very large write, then quieting down, then repeating - probably in time with the internal queue servicing.

Has anyone compared and measured these statistics, and if so is what I'm seeing normal for Linux POs? It doesn't appear to be impacting performance, but I don't know if it could be performing better if I can adjust this.