High disk utilization / low request queue and vice versa

[Jim McKee_1]

In the I/O by disk display in Glance I see some disks with high utilization and low request que (disk 7 in attachment) and others with low utilization and high request que (disk 20 and 22 in attachment).

I would have thought that these stats would go together. What would cause the situation I am seeing?

The server is a rx7620 running Oracle 10g and all of the disks are on an EMC DMX disk array.

Jim

[RAC_1]

There is no way for glance to know that the disk for which it sees high util is SAN disk, which in turn may be comprised of multiple disks. The applies to disk 20 and may be vice a versa for disk 22

[Kirby A. Joss]

Jim-

We saw similar stats when we tried moving some cooked oracle VG's from an EMC Symmetrix to a virtualized SAN. At the vendors recommendation, we collapsed a dozen or so PV's and LV's to 1 huge LV on 1 PV for the entire oracle DB. We saw the apps crawling, with low I/O wait and CPU utilization. Small block writes were worst. Best we could guess, we gave up too many buffers (allocated per PV and LV), actually blocking the app from running once the queues filled and the array's cache required posting. We solved it by adding PV's to the VG and breaking up the 1 LV to many.

-Kirb

[Suraj Singh_1]

The o/p suggests that some storage device is being used and LUNs have been presented to the HP-UX machine.

This behaviour is normal.

Rgds