Troubleshooting I/O Issues on HP-UX 11.31 - scsimgr Gyrations?

[Alzhy]

We're somewhat just getting deeper with 11.31 ona recently acquired rx6600. We hooked up an old Hitachi 9585V Thunder array followign all best practice guidelines and setting from the array vendor but we're getting horrible performance.

I see massive queuing as a symptom. The venodr recommends per LUN q depth settign to not exceed 8 but I am not quite sure where it is set in 11.31.

Any direction sirs and ma'ms?

I am still reading throuigh scsimgr.

[Steven E. Protter]

Shalom,

Define massive queuing?

sar output?

http://www.hpux.ws/?p=6

Perhaps gather and upload some data.

HP-UX 11.31 did make big changes in how it handles I/O.

Also, vendor best practice guidelines may not work for say oracle databases. Oracle recommends not using raid 5 for redo index and data.

SEP

[Alzhy]

HP-UX shalom10 B.11.31 U ia64 11/19/08

07:51:48 device %busy avque r+w/s blks/s avwait avserv

disk4982 100.00 335.89 259 17555 853.12 36.08
disk4983 100.00 582.95 453 17073 1143.59 24.09
disk4984 100.00 534.69 489 17557 823.33 18.41
disk4985 100.00 1037.61 325 10029 1746.10 36.97
disk5004 42.43 2.63 14 1759 75.25 170.37
disk5005 23.31 2.18 14 1759 32.47 91.01
disk5006 37.25 2.21 14 1760 25.01 103.33
disk5007 100.00 43.13 76 1629 225.14 97.69
disk5026 100.00 723.75 156 9052 1641.98 53.82
disk5027 100.00 1318.56 234 8735 1523.34 37.88
disk5028 100.00 302.57 421 15928 1043.70 25.30
disk5029 100.00 1552.61 322 9524 1176.55 31.80
disk5030 100.00 364.59 349 17415 757.57 37.00
disk5031 100.00 1260.28 301 12198 792.59 28.18
disk5032 100.00 1445.58 248 10486 1536.68 37.13
disk5033 100.00 1159.03 1124 34265 1993.43 12.46
disk5034 100.00 205.69 172 10301 963.93 51.00
disk5035 100.00 513.38 316 9632 1141.40 31.31
disk5036 100.00 525.72 319 9769 1314.15 33.06
disk5037 100.00 407.49 310 9638 984.50 29.18
disk5115 13.55 0.50 3 184 0.00 45.15


The array is an HDS9585V -- hooked up via 4x2Gbit FC Links. This server is the only client of the array right now.

The disks above are componets of volumes that are VxVM managed/DMP. The voluems are 4-way stipes and I am restoring data to them.

The disks above are LUNS carved from RAID7+1P RAID Groups on the HDS array.

[Steven E. Protter]

Okay then.

Good definition.

I would start with disk array configuration. There is not a heck of a lot you can do on the OS with a situation that bad.

Check the fiber card (fcmsutil) is working normally and connecting at a normal speed.

I'm thinking this has to be the array.

The performance documentation seems flawed.

See if you can use array utils to isolate the problem. If its heavy write action you know how to fix that, raid 10 or raid 1 LUNS.

Oracle archive log luns can be raid 5 because they are straight sequential writes.

SEP