LUN % busy

[Pedro Cirne]

Hi All,

I was wondering if someone can clarify if the metric %DiskBusy (SAR or Glance) is of any use when analyzing LUN devices from a disk array like Symmetrix DMX2000.

My DBA colleagues are arguing that we have a bottleneck on I/O because they can see devices (LUNs)100% DiskBusy (SAR or Glance)..but that's also true that avque, avwait and avserv are low:

Average c13t4d6 81.18 0.51 1150 68577 0.04 2.10
Average c12t4d5 68.40 1.54 1148 59799 0.77 2.49
Average c13t3d5 9.59 0.50 96 2041 0.00 1.06

Is %busy a good metric when analyzing IO bottlenecks on devices that are LUNs? I don't think so!

Thks

[Bill Hassell · Accepted solution]

100% busy means that during the measurement interval, one or more disks (or LUNs) were 100% busy. If this continues for an extended period, then the queue depth (avq) is useful to show the busiest LUNs. 100% is not any indication that the array is busy. You might have 500 LUNs and if only one is 100% busy, Glance will show 100%. The avq numbers reflect the request rate while avwait and avserv show the array's response time. It is absolutely normal to have certain LUNs busier than others. That's not a bottleneck unless the array's stats show the channels maxed out. It is very hard to generalize about the array's performance from the server side. Use the array utilities to look at the array cache and I/O rates.

[Pedro Cirne]

Hi Bill,

Thank you very much for your reply, I completely agree with you, I just wanted to have a second opinion and double check on this issue. I think the stats for avque,avwait and avserv are OK but I would like to have your opinion:

08:46:46 device %busy avque r+w/s blks/s avwait avserv
Average c13t4d6 52.22 0.54 777 37396 0.06 1.90
Average c12t4d5 56.40 0.83 963 27501 0.23 1.42
Average c13t3d5 14.61 1.97 95 8926 1.29 5.11

Thks!

[Bill Hassell]

> 08:46:46 device %busy avque r+w/s blks/s avwait avserv
> Average c13t4d6 52.22 0.54 777 37396 0.06 1.90
> Average c12t4d5 56.40 0.83 963 27501 0.23 1.42
> Average c13t3d5 14.61 1.97 95 8926 1.29 5.11

The last LUN looks fairly slow compared to the others -- wait and service times are much longer yet not as busy. I'd check the array's health just to make sure.

[Pedro Cirne]

Thank you