How to find out the actual physical path of pv . EVA preformace issue

[Nicholas Chang]

Hi All:
To improve IO adn enlarge disk space, we replaced a va7000 with a EVA40000.but the improvement is not as we expected.
and #sar -d has some avwait

sa$sar -d -f sa20080114.dat|more
device %busy avque r+w/s blks/s avwait avserv
c12t0d1 0.04 0.50 0 1 4.02 2.88
c12t0d2 86.03 0.59 864 5824 5.01 1.94


HP-UX 11.11
2 SAN 4GB switchs
2 2GB HBA
2 VD in EVA4000
2 VG in LVM
2VGs are both using /dev/dsk/c12txdx as primary pv path now.
Since we don't have Secure Path.is there any way to find out the actual PV device's physical path? then I can change one of the VG's primary path to operate these 2 VGs though differente EVA controller to share controller's loading.

OR , any other suggestion ? VD's prefer path of EVA ? any best practice of such case?

vgeva02
--- Physical volumes ---
PV Name /dev/dsk/c12t0d2
PV Name /dev/dsk/c13t0d2 Alternate Link
PV Name /dev/dsk/c20t0d2 Alternate Link
PV Name /dev/dsk/c21t0d2 Alternate Link
PV Status available
Total PE 3839
Free PE 0
Autoswitch On

----------------
vgeva01
--- Physical volumes ---
PV Name /dev/dsk/c12t0d1
PV Name /dev/dsk/c13t0d1 Alternate Link
PV Name /dev/dsk/c20t0d1 Alternate Link
PV Name /dev/dsk/c21t0d1 Alternate Link
PV Status available
Total PE 15357
Free PE 2221
Autoswitch On


----------------
Thanks

[Siju Vadakkan]

The diskname without "alternative link" in the "vgdisplay -v vgname" is the primary disk.

To make it alternative , just remove the primary and then again extend vg with the removed one. this will make alternative as primary and the primary will be added as alternative.

Can u pleae provide the disk size,

#diskinfo

[Nicholas Chang]

what I' trying to verify is which one shoule be replaced with to be the new primary path?
The purpose is sharing loading.

Ex: for the same VD in EVA ,if c12t0d1 is rp7410 HBA #1 to SAN switch #1 to EVA controller #1, then which path is rp7410 HBA #2 to SAN switch #2 to EVA controller #2?c13, c20 or c21?

the PE size is 16MB

thanks

[Siju Vadakkan]

Provide the following details:

# ioscan -fnkC fc
#diskinfo /dev/rdsk/c12t0d1

[Nicholas Chang]

sorry , I can not access to the devices right now

Do you mean I can track the connection form ioscan and diskinfo ?

I'll post ASAP

thanks a lot

[chris huys_4]

Increase scsi queuedepth of the implicated luns.

And see if that will allow to push through more IOs per second. When you do the sar, always do sar -d 1 1000.

Default scsi queuedepth on HP-UX equals to 8. (which is good to up to a 8Gbyte lun.) Youre c12t0d2 seems to be 64Gbyte, so scsi queuedepth of 64 should be better for this lun.

Check man scsictl to check how to set the scsiqueuedepth on a lun.

Then monitor the result, if it works, then you will also have to create a startupscript that makes sure that after every bootup the scsi queuedepth is set on the lun(s).

[Michael Steele_2]

Get some background data from these reports. As there are several ways to look at this and they all provide unique information.

Can you use 'ioscan -funC td' to find your HBAs. They'll be listed like /dev/td0, td1, etc. Then use 'fcmsutil /dev/td!' to find the WWN. Each HBA will have its own WWN. Also refer to the man page for additional arguements like 'stat'.

Re: "...what I' trying to verify is which one shoule be replaced with to be the new primary path? The purpose is sharing loading....".

This is a round robin algorithm that frequently comes up in the forum. Using 'vgdisplay -v /dev/vg##' you can list out all of your PV, primary and alternate.

Since you state "...2VGs are both using /dev/dsk/c12txdx as primary pv path now...."....

...then you will want to vgreduce /dev/vg02 /dev/dsk/c12t#d#....

Now your alternate slide up to be the primary and you 'vgextend /dev/vg02/c13t#d#...".

Here's what you have:

vgeva02
--- Physical volumes ---
PV Name /dev/dsk/c12t0d2
PV Name /dev/dsk/c13t0d2 Alternate Link
PV Name /dev/dsk/c20t0d2 Alternate Link
PV Name /dev/dsk/c21t0d2 Alternate Link
PV Status available

Here's what you'll get:

vgeva02
--- Physical volumes ---
PV Name /dev/dsk/c13t0d2
PV Name /dev/dsk/c20t0d2 Alternate Link
PV Name /dev/dsk/c21t0d2 Alternate Link
PV Name /dev/dsk/c12t0d2 Alternate Link
PV Status available

If trouble arise verify the PV with 'pvdisplay'. Refer to the -S autoswitch arguement.

"...y LVM is directed to automatically switch from the path it is using whenever a better path to the physical volume is available. LVM will switch paths when a better path recovers (after it had failed earlier), or if the current path fails and another path is available. This is the default....

http://docs.hp.com/en/B2355-60130/pvchange.1M.html

[Nicholas Chang]

thanks Michael:

I know how to change primary path.
but ,why c13? why not c20 or c21 ?

my purpose is to access 2 VGs through 2 totaly seperate fibre path and controller of EVA.

or is this only my imagination , nothing good for my avwait problem

And , I'll try to modify SCSI queue length.

Thanks a lot

[Michael Steele_2]

Well of course you can vgreduce c20 or c21. You can vgreduce up to three, maybe all four but I believe you have to keep at least one PV in the vg at all times. Its been awhile since I removed all disks from a vg.

vgreduce /dev/vg## /dev/dsk/c12 13 20 21
vgextend /dev/vg#3 /dev/dsk/c20 21 12 13

[Nicholas Chang]

thanks you all

I found that I can trace from the sequece of ioscan -fnk output.

c12 and c20 are through same HBA td0
c13 and c21 are throuth the other one td1

thanks Siji , I wish I had given you more points, or you can post another reply. ^^


thanks again

[Hein van den Heuvel · Accepted solution]

Nicolas,

The levels of usage you are currently should barely tickle the capacity of a single connection or the EVA. Don't woory about using multiple paths untill you approach 1/2 of theoretical throughput for longer durations.
So with 2gb fiber, you might start to worry at 50MB/sec, that would be 100,000 blks/s and 10,000 r+w/s.
Some might not like 1/2. So pick 1/4...
You would still have 5x room for more.

Chris>> Increase scsi queuedepth of the implicated luns.

Yes, it looks ( avwait -versus- avserv )like the throughput is constricted due to this artificial limit.

Chris>> Default scsi queuedepth on HP-UX equals to 8. (which is good to up to a 8Gbyte lun.) Youre c12t0d2 seems to be 64Gbyte, so scsi queuedepth of 64 should be better for this lun.

Good point, BAD explanation.
The desirable queue depth has absolutely nothing, zero, nada, to do with the lun size.

On of the most important factor the queue depth is the number of spindles behind the LUN presented by a controller like an EVA.
You should allow for at least one per spindle, more than 2 per spindle is unlikely to help much more.

Hope this helps some,
Hein van den Heuvel (at gmail dot com)
HvdH Performance Consulting

[Nicholas Chang]

Thanks Pharaoh

very clear explaination

this is what I wat to know

thanks again

[Nicholas Chang]

^_^