Poor perfromance after breaking and removing mirror root disk???

[Paul F. Bennett]

After breaking root mirror the system (at least vg00) is very sluggish. I noticed when running a sar -v it still shows the drive that was removed (c1t2d0) and shows it with values. Can anyone explain?

NOTE: ioscan shows NO_HW, as it should


#sar -v 3 3
07:50:26 device ~ %busy avwait avserv
07:50:29 c1t2d0 ~ 98.33 0.00 323.57 <--
c1t0d0 ~ 98.33 0.00 2608.15
~
~
07:50:32 c1t2d0 ~ 100.00 0.00 294.84
c1t0d0 ~ 100.00 0.00 3118.64
c2t2d0 ~ 0.33 0.01 6.79
~
~
07:50:35 c1t2d0 ~ 98.33 0.00 294.84
c1t0d0 ~ 98.33 0.00 2562.44
~
~
Average c1t2d0 ~ 101.44 0.00 304.10
Average c1t0d0 ~ 101.44 0.00 2760.54

[Paul F. Bennett]

Sorry about the format issue on the sar output.

The Average: %busy on the removed drive is 98.33, await is 0.00 and the avserv is 304.10

[Patrick Wallek · Accepted solution]

>>After breaking root mirror...

How did you do this? Did you do an 'lvreduce' to reduce the mirrors off of the disk, followed by a 'vgreduce' to get the disk out of the VG?

What does 'vgdisplay -v vg00' show now? What about 'lvdisplay -v /dev/vg00/lvol?' for some of the vg00 lvol's?

If the disk is still showing up in the LV's then that may be the problem.

[Paul F. Bennett]

I lvreduced the mirrors and physically removed the disks but did not remove it from LVM, in other words vgdisplay & lvmtab still shows it... but its only the root drive that slow access to the DB's that resides on vg01 are running fine

[Patrick Wallek]

Try reducing it out of the VG.

# vgreduce -f vg00

Will force a reduce of missing PV's out of the VG.

[Hein van den Heuvel]

Yes, those avserv time are horrible, but what is the access rate?

Is the system experiencing any actual performance issues, or are you just concerned about an ugly number on a report?

Could it be that from an HPUX perspective the system is 100% busy waiting for long times for responses which never come as the drive is gonzo?

Also, given a high read activity on VG00, some slowdown should be expected when removing a mirror member. From the "HP-UX Performance Cookbook":
"Sometimes for read-intensive workloads, mirroring can improve performance because reads can be satisfied from the fastest disk in the mirror"
Ergo: un-mirroring can decrease performance.

hth,
Hein van den Heuvel
HvdH Performance Consulting

[Paul F. Bennett]

I just realized that there was a small lvol on the failed disk, that was NOT mirrored. I'm thinkning this may be partly causing the poor performance... but when I attempt to lvreduce it it fails, it basically says it cant access the drive. I think I will replace the drive and hopefully the vgcfgrestore will allow me to lvreduce it or mount it. We'll see :-)