[sbzx]

Hi,

We're setting up a 55A plus HMC/APV with eight internal disks (2x73GB, 6x146GB). VIOS is installed on the two 73GB disks, and the rest are allocated into two groups of three. The two groups will serve the rootvg disks of six LPARs. We created the two VGs, using "mkvg -vg client1vg hdisk2 hdisk3 hdisk4" and "mkvg -vg client2vg hdisk5 hdisk6 hdisk7". So far so good. Now, when we created the LVs to be virtualized as rootvg disks ("mklv -lv root1vg client1vg 50G hdisk2 hdisk3 hdisk4") we noticed that the LVs had been created with inter-policy minimum on only one disk, and no way of changing that to maximum by padmin. Root can of course do that. In fact, this is exactly what we did. Went to "oem_setup_env" and did the LV creation from there using the "-e x" switch with mklv.

So my question is this: is there any recommendation by IBM to discourage using the maximum inter-policy on such LVs? What sense does it make to restrict standard VIOS mklv in such a severe way? And will my system be supported/are there any risks involved when creating LVs using root instead of padmin?

Thanks.

[Mark Taylor]

As a rule of thumb you have to take into consideration that commands outside the padmin rksh are not supported .. if you cannot do it with a padmin command then you should not be doing it.

I take it that you have 2 x storage vio servers and you are going to mirror at the client o/s level ?

Rgds
Mark Taylor

[sbzx]

Hey Mark, thanks for the reply. No, I have only one VIO server with two VGs for VIO clients. Clients will mirror between the two.

What triggered my original question is that it makes little sense having multiple disks in a VIOS VG and not being able to distribute an LV across all the PVs in the VG. I find it quite strange. Since there will be 6 client OSs, I expect heavy load on the disks due to possible paging, so I think maximum inter-policy would be most beneficial.

Thanks.

[Mark Taylor]

You are asking from a performance perspective right ? and yes, i have come across this before where I have wanted to do a "poor mans stripe" to balance the load across more than one spindle but it is not supported as you cannot do this with a padmin command.

I cant see why it wont make it into future code though.

You could instead of just 1 x LV across 2 disks actually have 2 x LVs (one on each disk) and pass them both up to the client, then migrate / spread your client LVs across the 2 x disks ((presented LVs)) and effectivley you have achieved the same setup.

HTH
Mark Taylor

[sbzx]

We have made a decision and we're going with the unsopported version for now. It's a test phase only at this time. We'll have our backups ready so we can revert back to the supported and presumably slower setup in case of an unforeseen error.

By the way, having given it a little more thought, to spread an LV across 3 disks can be achieved in two ways with either maximum intra-policy (poor man's stripe) or minimum intra-policy (with a large enough LV). I don't see the two very different from an "underlying OS vs. VIOS" point of view.

Thanks for your thoughts, Mark.

[Mark Taylor]

>>I don't see the two very different from an "underlying OS vs. VIOS" point of view.

only in the fact that you will have very little control on the size of the LV .. i.e. you need to make it big enough so that it wont fit on 2 PVs so that it will spill over onto the 3rd and in fact to get it even you will have to make it the size of 3 x PVs ... which you can only pass up to 1 x client.

Rgds
Mark Taylor

[sbzx]

Of course, you are right, but that is not what I mean. I mean VIOS interacts with an LV through LVM and AIX. VIOS probably has no direct connection to the raw LVs so it should not care how the LV is actually distributed across the disks, regardless of LV size. Viewed from VIOS an LV is just an LV, its layout should be irrelevant.