iostat - how accurate is it?

[Mott Given]

We use a capacity planning tool from a vendor who shall rename anonymous to study applications running on HP systems. The vendor creates their disk stats for the analytical models that we build by using HP's iostat. The anonymous vendor maintains that their disk utilization calculations are accurate regardless of:
1) where the I/O was actually serviced (ie. physical disk, cache on disk, cache on the controller)
2) it does not matter whether the device is RAID or non-RAID
The anonymous vendor has their own collector running on the HP systems which samples system data every 10 seconds, including the iostat type info.
Does the accuracy of iostat and the architecture of current disk devices support the vendors 2 claims above? I would have thought that whether the I/O was a physical or logical I/O would have had more of an effect on disk utilization.

Mott Given

[Sunil Sharma_1]

Hi,

Data collected through iostat or some other tools which work in same way may not give you realistic information if you are using hardware RAID.

Sunil

[doug mielke]

iostat is accurate, as are the sar -d and -b commands that you could look at.
Their claim of accuracy is valid as well. It doesn't matter RAID type, external cache, etc.
But, all that you measure at the OS level is service time. The OS, and most tools that run on it, have no way of knowing if the service came form a cacheing controller or hard i/o.

To help limit the effect of caching, about all you can do is try to ensure that your disk reades are not repetitave, and scattered around the disk.