Need to find a way to force array utilization lower

James,

I assume that you are using NST... It is "only" bothered about the IO on the FLARE spindles. In most cases a single RAID-5 RG covers the first five (5) spindles. Hence you need to drive down the IO on the LUNs on those spindles...

Obviously you will want to do your upgrade during a quiet period anyway.

I hope this helps
Carl

Interesting. I am also preparing a FLARE upgrade (since several months...). I have just added some DAE so I have space available and I could migrate LUNs from the first Raid Group to new disks. Is it advised ?

Hi Rob,
So I am going to migrate luns But leaving 5 disks empty is quite expensive ! I think I will re-use them after the upgrade.
My NDU is from 22 to 26.

It would indeed seem so
Furthermore: if you upgrade from F19 to 26, you need more free space on the Vault drives since F26 needs more space....
What we do nowadays is to install a Clariion with the smallest drives possible in the first 5 positions and then leave them empty for any future Flare upgrades.

Re-use them with LUN's with modest I/O, like a fileserver or something or to store your precious ISO's or VMware templates

But if you need to empty them for extra space, I'm not sure of since F22 / 26 don't differ (much) I think. Better ask your CE if extra space is needed if you go from 22 to 26.

Carl,
I don't think that is the only checks it performs. Remember that during a FLARE upgrade the SPs will reboot one at a time so the CPU utilization on each SP needs to be below 50% to ensure that one SP can host all the i/o and disks during a reboot of the peer SP.

Aran

Aran,

OK very valid point! The checks are obviously there for valid reasons. I should have formulated my answer more carefully as the original question was to find ways to force array utilization lower.

I shall stand corrected
Carl

Or sit , I find it hard to type standing...

But yes the question itself is a tough one. Without some cooperation from the administrators of the applications hosted on the array it will be difficult to minimize the utilization of the array.

James,
Do you not have a maintenance schedule for systems in your infrastructure? A FLARE upgrade while usually non-disruptive does have the potential to impact not only i/o but there is the possibility of an actual outage if something goes wrong.

FLARE upgrades should be done during low i/o periods after full backups of all the systems on the array. Because the SPs reboot during the upgrade write caching will be disabled for most of the upgrade process which will seriously impact the performance of the array.

Aran

R26 does not require more reserved space than R19.
There are differences in reserved space size, but it is platform dependent (depends on array type), not FLARE revision specific.
For example; a CX700 will require/use the same amount of reserved space, independent of whether it is running R16, R19, R24, or R26.
A CX3-series array requires/uses a larger reserved space - this is true independent of whether it is running R22, R24, or R26.

Regards,
DGM

I must say I've never did an upgrade myself, but that you mention it, the extra space I was thinking about might have to do with an upgrade from CX to CX3. In that perspective my answer as well as yours makes sence

Yes, an upgrade ("platform conversion") of a CX-series array to become a CX3-series array, would involve making sure there is enough space available on drives/LUN in the vault locations, to accept the change in reserved space size.

Regards,
DGM

I haven't played around with Navisphere QoS so I don't know if that will achieve the desired results. You could at least try it out on a few hosts to see if it works for your goal before implementing it on all hosts for the upgrade.

Have you looked at Analyzer to see if there is any time during the day/week that the array is at a lower level of utilization?

NQM Is pretty nice but still needs some work IMO.

It's hard to manage multiple things at one.  If you've got a few problem areas, then it's pretty good.  I think it would be a pain to configure for your whole array.

I got NQM to manage some of our new Sun servers.  They push writes too fast (super fast servers) and fill up cache to 99% with anything fewer than like 60 spindles (RAID10) for a LUN.  NQM let me limit the IO of those LUNs so that the disks could keep up.