Autoraid 12H Performance Question

[Lothar Krueler]

Hello,

one of our 12H goes slower and slower. Weekly i have to run a big copyjob (removing old data before) to this 12H. First time after creating the LUNs the job takes approx. 5 hours. Each time it runs it takes much longer. Now it's nearly 16 hours. If i recreate the LUNs it will start with 5 hours again. Because this 12H is low in space, thought of heavy balancing between raid level 0/1 and 5, so i hoped that i could reduce that by allocating all avaible space to LUNs for forcing raid 5. It didn't help really.

The VG is striped across two LUNs, using the X-Controller primary for the first LUN and alternate for the second LUN.
Last week i configured sar for collecting performance data. I wonder why the values for "avwait" of the two devices (LUN1 and 2) are much different. I expected them for beeing approx. equal.
The output from sar, arraydsp and vgdisplay is attached in a *.txt File.

Why do the values differ so much ?

Thanks in advance
Lothar

[Mark Grossman · Accepted solution]

Lothar,
It has been our experience that without a lot of free, unsued, virgin space available, the autoraid will just not perform very well.

As you've seen, just deleting files will not put that space back into the unused mode. You have to totally recreate the disks to make that clean space usable by the system.

You will drive yourself crazy trying to run statistics on that autoraid if its nearly full.

My humble advice would be to either add more or bigger disks if possible, or look for a newer technology array.

Mark

[Lothar Krueler]

Mark,

thank you. I think that changing disks to bigger ones would result in better / normal performance, too. My question is, could it be, that one of the LUNS is much slower than the other one ?

Lothar

[Mark Grossman]

Lothar,
not sure whats going on with that sar output. However, one of the luns is much much larger than the other and could account for the difference in statistics.

Mark

[Lothar Krueler]

Mark,
thank you again. The LUNs are exact equal in size, each 10771 PEs - but they seem to be not equal in performance. It looks like the LUN presented by c9t0d0 is much more slow than that presented by c10t1d1 - with nearly the same busy-values.



Since there are no further responses, i think possible answers to this question may only have an theoretical charachter, that really will not be of any use.
Thanks again

Lothar