sar -d

[pa8700]

hi all ..I have attached a sample sar -d out put. I can see some disks have more disk que like above 4 and the service time is less ok like 14 .

But some disks have less que 0.5 and the service time is big like 32 .

So which will be considered as a disk bottle neck.

[Pete Randall]

From the Performance and Tuning class I took many years ago:

"If the %busy is greater than 50% or avwait is greater than avserv, it indicates we have a disk bottleneck."

Maybe that helps.


Pete

[Tim Nelson]

One thing to note.

A number of disks you are looking at have NO r+w/s or blks/s.. With no IO going to those disks the queue and service times are irrelevant.

I initially expect that c2t0d0 and c0t0d0 are some locally attached slower single spindle scsi and the service time is average for those type.

Any disk servicing low ios and small blocks will suffer when competing with larger sequetial ios..

Notice c31t15d3, no ios, 20 blk/s waiting a long time for service. poor little guy, but that is the way it is. ( he is probably asking himself, "where in this vast void is the little block of data I am looking for". Then again it could be a performance anomoly because there is little perf data to work with.

I may be off but I see a whole lot of nothing going on here.

and.... this is only one snip in time.. let 'er run and monitor over a period. snap shots can be misleading.

[pa8700]

what about the que .i have attached the avarage of daily over 5 days..
For me I hope there is a bottle neck on there local vg00 disk.So the performance of the server is going bad.

Its a superdom...

Please let me know your comments

[pa8700]

see the attachment

[Tim Nelson]

Again, the ios are only 30 io/s...

what is on c2t0d0 and c0t0d0 ?

Just the OS ?
What else is housed on these disks ?

We may have to take a couple steps backwards as from what you have shown there is little if nothing going on. You cannot troubleshoot performance when the server is idle but I think the picture is incomplete.

What does average CPU util look like ?
What does mem util look like ?
What type of application are you running?

[pa8700]

hi these local disks host vg00 ,including OS and some other filesystems like /oracle , /tvoli ,/bmc , /var/opt/perf/datafiles.

again thsis server is a MC/SG node

[Tingli]

Only c2t0d0 and c0t0d0 have issues although they are vg00 only. It is a little bit strange this way and need to find out what is running on them to cause it.

[Torsten.]

Are the disks in a DS21x0 by chance???

[pa8700]

these are the internel disks of a sd64 superdom

[Torsten.]

A Superdome has NO internal disks ... never.

[pa8700]

You are right .But may be on the expansion cabinet .see the below out puts

$ uname -a
HP-UX hpx159 B.11.23 U ia64 0593007228 unlimited-user license
[psivanan@hpx159:/home/psivanan]
$ model
ia64 hp superdome server SD64A
[psivanan@hpx159:/home/psivanan]
$ ioscan -fnkC disk | more
Class I H/W Path Driver S/W State H/W Type Description
===========================================================================
disk 0 0/0/1/1/0.0.0 sdisk CLAIMED DEVICE HP 73.4GST373207LC
/dev/dsk/c0t0d0 /dev/dsk/c0t0d0s2 /dev/rdsk/c0t0d0 /dev/rdsk/c0t0d0s2
/dev/dsk/c0t0d0s1 /dev/dsk/c0t0d0s3 /dev/rdsk/c0t0d0s1 /dev/rdsk/c0t0d0s3
disk 1 0/0/1/1/0.2.0 sdisk CLAIMED DEVICE HP 73.4GST373207LC
/dev/dsk/c0t2d0 /dev/rdsk/c0t2d0
disk 4 0/0/1/1/0.4.0 sdisk CLAIMED DEVICE HP 73.4GST373207LC
/dev/dsk/c0t4d0 /dev/rdsk/c0t4d0
disk 2 0/0/1/1/1.1.0 sdisk CLAIMED DEVICE _NEC DVD+RW ND-2100AD

[pa8700]

HI torsten.

Is it avvisable for superdom to have OS on the sigle spidle disk . Or there is huge deference if it is on a FC array.

[Torsten.]

The DS21x0 is invisible in ioscan.

Have a look!

However, please post a complete

# ioscan -fn

[pa8700]

please find the attached out put of ioscan .
is there something wrong with DS2100.

[pa8700]

ioscan out put

[Torsten.]

Nothing attached.

[Torsten.]

I would like to see

# scsictl -c get_target_parm /dev/rdsk/c0t0d0

# scsictl -c get_target_parm /dev/rdsk/c0t0d0s2



and

# scsictl -c get_target_parm /dev/rdsk/c2t0d0

# scsictl -c get_target_parm /dev/rdsk/c2t0d0s2

[pa8700]

Torsen pls find the attachment

[Michael Steele_2]

Hi

a) you have a disk bottle neck when avwait > avserv. Only one metric meets this condition: c31t15d3.


The DS2100 series of SCSI bays are standard to superdome O/S disks. They are not internal per se, and they will be about 15 feet away SCSI cable connected. Here is where you will find you O/S drives.

Note: Since you can see the disks, there is nothing wrong with the DS2100. Also note that besides a back plane, slots and drawers for your disks to slide into, there is nothing that shows up in ioscan.

Again, if you can see the disks in the DS2100, then the DS2100 is working fine.

[Torsten.]

I expected a different controller, but this looks fine.

[pa8700]

you mean the there is no bottle neck here on the local disk

[Michael Steele_2]

Hi

I see consistent disk bottlenecks for c2t0d0. This is not unusal. This is the O/S disk. And unless you have some non-O/S related file systems to remove out of vg00 and into another vg, this is life.

Note that it is also life fo 99% of all boxes and that you are not alone.