I/O performance tuning

[Lynn Hsu_2]

Hi,
I need some help with disk i/o tuning. Our
server is L3000/HPUX 11.11 using PROGRESS
data base with AI (after image) turned on. I noticed whenever a cron job copying
data from one LUN to another (we use EMC Clariion Cx500 disks and was told it has 400GB cache), glance+ shows disk i/o "alert". The fstab line is like this:

/dev/vg12/lvol1 /proolhist vxfs rw,suid,largefiles,delaylog,datainlog 0 2

Anything can improve on this line? Are there
Kernel parameters can be turned to optimize
disk I/O? I'd appreciate any input...

Thanks,

Lynn

[A. Clay Stephenson · Accepted solution]

You can pretty much ignore this warning. Any host-based performance tool (Glance, iostat, sar) can only know that a tremendous amount of i/o is going through what it sees as 1 disk -- that's all it can know. The "one" disk may actually be comprised of many physical disks + cache inside the array but Glance doesn't know this.

You can make things APPEAR to be better by breaking up a VG into smaller LUN's but the actual throughput will likely be unchanged. You can actually improve i/o by breaking VG's into as many PV's as you have separate i/o paths for and then striping each LVOL in this VG across all PV's.

[Steven E. Protter]

Shalom Lynn,

First confirm a problem exists.

Measure stats:

http://www.hpux.ws/system.perf.sh

SEP

[Carlos Roberto Schimidt]

Hi Lynn,

See this link where I attach a perfomance cookbook.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1009131

Read about mount options in VxFS filesystems. i.e:

mount -F vxfs -o largefiles,mincache=direct,convosync=direct

Schimidt

[Lynn Hsu_2]

Thanks for all the replies. I also read the
cook book. It's very practical.

Lynn

[Lynn Hsu_2]

good info. Thanks!

Lynn