Time onto conclusions
Thing which it was been possible to make in system (on host), which sure they will improve bit performance:
1) Change following parameters of file systems (vxfs) in accordance with recommendations (
http://docs.hp.com/en/B3929-90011/ch05s07.html and
http://h71028.www7.hp.com/ERC/downloads/4AA0-2030ENW.pdf) - in this case: read_pref_io=1024Kb, read_nstream=4, write_nstream=1. Make this online across vxtunefs command or offline across file /etc/vx/tunefstab.
2)Change mount options for oracle file systems (OnLineJFS options) in accordance with recommendations
http://h21007.www2.hp.com/dspp/files/unprotected/database/HP3KOracle.ppt. This sure improves the buffers utilization.
3)One should to exchange place of some files (.dbf .ctl ...) between file systems (lv and pv)so to discs were similarly busy. On present results one disc c32t8d2 is busy.
Thing which it was been possible to make EVA and methodology, which sure they will improve performance:
4) To minimize impact of snapclone performance on system performance, transition cache on source virtual disks to write-through mode using Command View EVA before starting the snapshot or snapclone. The impact of creating one snapshot may be tolerable and may not require this action.
However, creation of multiple snapshots of the same virtual disk requires this action (Doc best practices for EVA - white paper).
5)If you are also taking a backup at the same time, you may want to move it until later. The snapclone actually does a snapshot until the snapclone process has completed (copying data over).
6) Take EVAPerf and analyse situations during snap clone and without.
Before of any changes execute:
#timex dd if=/data/ALPHA/dbs02/some_big_file of=/dev/null bs=4096k count=10000&
#sar -d 5 10 > before.log
#sar -b 5 10 > before.log
and after changes make the same to compare results.
Regards LiPEnS
