What are acceptable sar -d values for system performance?

[Ben_40]

I have average sar -d stats as follows:

device %busy avque avwait avserv
c0t1d0 3.93 1.94 3 34 17.28 34.93
c1t4d0 2.87 2.04 2 28 16.16 33.90
c0t2d0 15.16 20.73 14 204 186.40 51.99
c1t5d0 14.42 31.33 11 167 364.26 78.15
c0t3d0 7.30 3.67 7 103 32.02 26.71
c1t6d0 7.89 3.94 7 98 45.51 35.08

I read on these forums that I may have a disk bottleneck if any of the following occur:
1. %busy > 50%
2. avque > 0.5
3. avwait > avserv
4. avwait > 6
5. avserv > 1 (!)

I got these rules of thumb from other forum posts and would like to verify that this is correct. Also, do the stats that I listed above indicate that I have a disk bottleneck?

[Sridhar Bhaskarla]

Ben,

Theoritically yes. But I don't care about any of the things if my %busy is less than 80 and if %wio is less than 10 in sar -u.

Your case is interesting. You are using some devices less than 20% but yet you are seeing avqueue > 0.5. This might be because there are a lot of long sequential writes?. And the formula %avserv > %avwait is applicable only for reads. I have been seeing on my systems that %avserv far below %avwait but without any queue.

So, I am more worried if my %wio in sar -u is > 15 + %avqueue > 0.5 + %busy > 80 + avwait > avqueue.

Please be noted that I used + above means all the conditions must be true to say that I have disk bottleneck.

By the way, what does your sar -u look like along with sar -b ?

-Sri

[Praveen Bezawada]

Hi
Your %busy is ok... though 50% is considered a bottle neck we have applications running with around 65%.
Another way to identify a disk bottleneck is to use "avque >= 2". You seem to be having a problem here.
Check the values of %rcache and %wcache, if the values are less than 90 and 65 respectively,it means that buffercache it too small.
Better use glance to monitor the performance, it gives a bigger and a better picture.
Hope this helps...
...BPK...

[Manel Ventura]

Hi Ben,

I think so. I have compared your list of response times with my own list and I find high times of response. However, it depends of the type of disks, configuration, load...

For an EMC Symmetrix with disks of 36Gb in stripe-4, with 450 users, the result should be like this: (see attach file).

[Sanjay_6]

Hi Ben,

Look at this thread. It might help in what you are looking for.

http://us-support3.external.hp.com/cki/bin/doc.pl/sid=5bb27d1a14db068f31/screen=ckiDisplayDocument?docId=200000049436468

Thanks