I would like to see if anyone has previous experience and/or a strong opinion one way or another concerning design. I am running two pSeries hosts with AIX 7.1 clustered using PowerHA. Lets say I have (20) EFD's available on the symmetrix. I use a R5 3+1 and then stripe those together to make a large meta-lun. When presented to AIX this gives one hdisk where we assign a volume group, LV's and file systems. Time passes and all is good but we are eating up this space. We purchase another (20) EFD's. We now have two options.
Option 1) Create additional R5 3+1's and stripe those against the existing meta.
Will I hit a drive limit?
Will this cause a major performance headache during the re-stripe for the current hot file system?
Will the AIX cluster handle growing the VG without an outage?
Are there other considerations?
Option 2) Create a new striped metalun with R5 3+1's (same as the original set) and add that to AIX. This will add another hdisk. Essentially this will be host based concatenation (rather than on the Symmetrix).
This will not stripe data but since these are EFD's do we really care?
Are we creating any sort of a redundancy issue or data integrity issue now or later down the road by having data blocks randomly spread across these hdisks?
Will the AIX cluster handle this without an outage?
And another option would be to make the devices VP devices, then you don't need to worry about the backend striping. When adding new disks, just add them to the pool and re-balance.
Quincy - adding the disks on the storage side is not really the issue. It is expanding the VG that is untested.
Dynamox - I agree it provides more I/O queues in the OS. But AIX will do a concatenation. Do you feel this is find since we are on EFD's? I am not sure I would recommend that for spinning disk but maybe in this context it is just fine. It just seems to go against my better judgment.
You can always stripe in LVM on AIX. Have you considered creating a pool like Quincy56 suggested ..easy to add capacity in the future and keep it simple on the host side? Later on if you decide to implement FAST-VP you will be well positioned.
Quincy561
1281 Posts
233
0
Posted December 10th, 2012 09:00
And another option would be to make the devices VP devices, then you don't need to worry about the backend striping. When adding new disks, just add them to the pool and re-balance.