Here is a story I’d like to share with your, my dear, reader. Couple days ago we finally found a remedy for a support case opened in Sun, which even a support engineer christened as “strange”. The problem was initially discovered whilst the root disk mirroring process was on-going. I noticed that it took flagrant amount of time to complete and once it was over I used dd and iostat tools to recreate the problem. And here what I found:
# dd if=/dev/rdsk/c10t0d0s2 of=/dev/null bs=128k & # iostat -xnzM 5 extended device statistics r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device 20.4 0.0 2.6 0.0 0.0 1.0 0.0 48.9 0 100 c10t0d0 extended device statistics r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device 20.4 0.0 2.6 0.0 0.0 1.0 0.0 48.9 0 100 c10t0d0
Exactly the same output I got trying to write to the disk, so the issue was obvious – 2.6MB write/read was abysmally low even for D240 JBOD. Btw, our configuration was the following: SF6900 with Qlogic Ultra3 SCSI ISP10160 connected to D240. Whilst working on the problem we’ve tried to do many things:
- Fiddle with system parameters, i.e.sd_max_throttle, and tried to switch from full to split bus configuration. Didn’t help.
- Took another Qlogic Ultra3 SCSI ISP10160 – the problem was still there.
- Connected D240 to a different HBA and couldn’t reproduce the original issue. So it looked like D240 wasn’t a perpetrator.
- Plugged D240 back to the original HBA and tried to swap the slots in IO board – same issue.
- Upgraded the disks’ firmware to the latest version (116370-15) – the problem persisted.
In the end the support engineer ordered a slightly different disk model than we originally had. Our Seagate (ST373307L) were replaced with ST373207L and the problem had finally disappeared.
extended device statistics r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device 291.4 0.0 36.4 0.0 0.0 1.0 0.0 3.4 0 98 c10t0d0 extended device statistics r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device 291.2 0.0 36.4 0.0 0.0 1.0 0.0 3.4 0 98 c10t0d0
Hazy miracle I hate. We still haven’t received any technical details but were promised to have the needed information as soon as it’s available from the PTS engineer situated in Sun European Competency Centre.