Showing results for 
Search instead for 
Did you mean: 

sar -d avserv times

Go to solution
Super Advisor

sar -d avserv times

We have been collecting some sar data for our disks that are attached to the SAN. I've formatted the sar -d output so it can be sorted in a spreadsheet and I'm seeing the following:

9:00:00 device %busy avque r+w/s blks/s avwait avserv
9:40:30 c10t0d2 68.49 0.5 5 128 0 681.01
9:28:10 c10t0d2 39.22 0.5 7 189 0 205.15
9:27:50 c11t0d2 58.04 0.5 6 158 0 139.11
9:39:30 c18t0d2 54.2 0.5 7 210 0 131.61
9:32:10 c18t4d5 49.55 0.5 6 190 0 128.45
9:40:20 c11t0d2 40.3 0.5 7 197 0 102.28
9:40:10 c18t0d2 55.44 0.5 7 230 0 85.14
9:39:50 c10t0d2 44.6 0.5 6 186 0 79.21
9:40:30 c18t0d2 7.88 0.5 6 164 0 76.36
9:21:40 c12t4d4 1.8 0.5 4 62 0.56 60.5
9:40:30 c12t0d2 7.58 0.5 8 182 0 59.18
9:32:00 c11t4d5 30.97 0.5 5 169 0 58.27

Our HP-UX 11.23 system is connected to an EMC DMX-3 and I'm trying to get a good guage on what exactly is considered a high service time. From the looks of what I copied into this inquiry, it's looking like we have some seriously high service times.

Is it fair to say anything > 10ms is not very good? We have tier-1 storage on Raid 1 (mirrored) protection in our situation here.

Any help or guidiance would be greatly appreciated.


Duncan Edmonstone
Honored Contributor

Re: sar -d avserv times


A simple rule of thumb is:

<10ms = good
<20ms = acceptable
>20ms = potential performance issue

So for these disks here, something is definately not right!

The worrying thing is that the disks don't even look that busy in that none are anywhere near 100% busy during the intervals and there aren't a great number of IOs per second either.

That said, you're obviously not showing us all the facts here, as this appears to be a list sorted on the worst service times. More interesting would be the general trends and averages rather than just picking out these peaks.




Super Advisor

Re: sar -d avserv times

Thanks Duncan,

Like you mentioned this is not merely all the facts, just a quick snapshot of some of the worst observations throughout the course of an hour slice. We have way too much data to really take to one of these forums. The rule of thumb that you mention is meaningful and helpful though and I think if we just look at that alone we can confirm we have some issues here that need to be further investigated. I will take this further into the SAN and look there to see if we see any "hot devices" and start there.

Thanks for your input...
Super Advisor

Re: sar -d avserv times