dba needs sysadmin help on optimizing configuration of array/ctrlr for db usage
Hello to all and I hope you can help my confusion. I want to reconfig the array but want to be sure I am not going to toast the vg00, lvm config, or raw spaces. I inherieted the current system: HP 9000n800 11.0 HP 12H Raid fully populated with 12 9 gig drives (No unallocated space on the 12H, so I believe it is operating at Raid 5 for EVERYTHING!) 4 LUNS 2 ctrlrs one active one spare Informix 7.3 on raw dbspaces vg01 has the app executables and data files vg02 raw dbspace and tempdbs vg03 has raw test environ
Goals: -Modify the controllers: both active for 2 LUNS each and spare for the other 2 -Fragment the db into various dbspaces and map them to diff VG's, if that will improve perf (I don't think it will, since the 12H manages everything using it's own algorithyms) -Drop one or more unused dbspaces (8 dbspaces allocated < 10% in actual use) -Create additional tempdbs in diff VG -Most important: drop and unallocate the excess dbspaces/raw disk and give it back to the 12H so it can re-build the Raid 01 capacity.
Q: -am I off base in any of the assumptions or goals? -will changing the active controller paths to LUNS do what I want for the raw spaces, and is there anything I need to do besides the vgreduce/extend for the raw files and links? -what caveats are there besides be very careful when I kill off the excess dbspaces and...yikes...delete the logical drive.
tia to any who will respond!
Doug p.s. what is the surest method to be sure I am correct in the assumption I have no unallocated disks/space in the 12H?
Re: dba needs sysadmin help on optimizing configuration of array/ctrlr for db usage
Wow this is a big job. First things first. Make a recovery tape of your OS. You can use ignite make_recovery for this. Secondly shutdown you databases and do a cold backup of everything. If you do toast everything recovery should be easy from here.
Now comes the planning stage. You say you have two paths with each one having an alternative path. Try and plan your VG's around this so that your load will be balanced. Personally I like working with LVM rather than raw volumes it gives a bit more flexibility but of course there are instances where this is not possible. If you are thinking about splitting dbspaces across VG's make sure the VG's are split across the controlers. Maybe even use physical VG's
Re: dba needs sysadmin help on optimizing configuration of array/ctrlr for db usage
I guess no matter how you fragment your db or re-configure the array controller paths that won't give you noticeable performance gains. The effective change would be on reducing the array utilization and hope that the 12H could operate in mirror mode instead of RAID 5. We did some in-house test before on the same 12H array and found the performance difference was quite significant.
Make sure to have a good recovery plan before wiping out the array.
Re: dba needs sysadmin help on optimizing configuration of array/ctrlr for db usage
Hi Doug,
The above comments about backup and recovery I would most certainly do. We're using a 12H with Oracle but are not using raw devices. From what I know the performance gain is only about 2% and this is to little for us to give up the more convenient LVM-administration. As far as your questions are concerned. 4 LUN's is excellent. You can switch primary and alternate with the system online using vgreduce and vgextend. If you can free up some diskspace for RAID 0/1 that would be good. However, the 12H will ALWAYS put data in RAID 0/1. Use the arraydsp -v command to see how big it is. If there is a lot of activity, you will see it grow larger provided the system is not completely filled up. I use a simple script writing this info to a log-file every hour. With arraydsp -m you can check you write working set and other performance metrics of the AutoRAID. Check these regularly to notice the development of your system. Using both RAID-controllers will give you much more performance gain than anything else probably. In Oracle, more tablespaces are good for performance. Using more VG's on the 12H does not make any difference in my experience since everything is striped over all disks. If you have redolog-file on the 12H, consider moving them off the 12H to another disk because they are written to constantly which will keep to internal SCSI-queue busy.
