HPUX with EMC Clariion performance query

[Adam Noble]

All,

I was hoping somebody may be able to assist. We have a hpux 11.00 server connected to an EMC Clarrion. We have been experiencing some performance issues and we are unsure if they are related to disk. EMC have stated that we should not stripe our filesystems on the host as striping takes place at the back end.

I believe this causes us some issues trying to determine if we are suffering performance degredation due to I/O performance. We have a logical volume which contains 4 LUNs presented to the server from the Clariion. At the back end these LUNS are spread across many different disks.

The database tables reside within this logical volume however obviously some are hit more frequently than others. This means that quite often one particular LUN is being hit regularly and Glance/Perf etc seems to suggest it is 100% utilised. Other metrics such as disk queue and Average service time suggest to me that actually all is fine.

EMC themselves are saying that the utilisation metric within glance is now not relevant, so therefore how can I determine if we have an I/O bottleneck?? Does anyone know how glance determines the utilisation percentage of a disk?

[RAJESH GANGADHARAN]

Hi Adam,

Try with "sar" with "-d" option. Since sar used across all platforms of Unix.

for example, sar -d 5 1000.

Thanks.RGP

[RAJESH GANGADHARAN]

Hi Adam,

Try with "sar" with "-d" option. Since sar used across all platforms of Unix.

PLease note the high disk activity on the disk and also make a special note that avwait should not be greater than avserv for a disk which is showing high activity.

for example, sar -d 1 10
05:35:06 device %busy avque r+w/s blks/s avwait avserv
05:35:07 c100t8d0 5.94 0.50 6 30 8.65 15.67
c100t10d0 4.95 0.50 4 22 8.32 14.69
05:35:08 c100t8d0 4.00 0.50 4 18 4.25 9.98
c100t10d0 2.00 0.50 2 10 3.58 10.35
c0t8d0 2.00 0.50 2 20 1.24 9.60
c1t8d0 3.00 0.50 2 20 5.34 10.37

Thanks.RGP