Disk io problem

[Rene Mendez_4]

Hello
Configuration:
LVM
HPUX 11.11
Storage Hitachi FC
File System on the luns hitachi

The problem is the disk io rate
if check with iostat show 30Mbs rate, sar -d show 100% use

When use dd command the rate is 120Mbs,
When use cp file system to file system the luns the rate is 30Mbs

I installed OVPM: The disk pick utilization is 100%
The disk rate is 30%

I modified the dbc_max_pct=15, dbc_min_pct=5, after 5;2, but the problem is same.

How to determine the problem with IO, ?

Regards
Rene

[nanan]

Hi Rene

I'd like to tell you there is a probability of difference between dd and file copy

because file copy need more procedure in Kernel, like check inode and check where is etc. but dd do not
and if you are going to cp a filesystem which is consisted in so many tiny files, it would be a cause of low speed in disk IO

Regards
nanan

[A. Clay Stephenson]

First, you should understand that performance tools like sar and Glance don't know that this "disk" is actually an array LUN that might actually be comprised of many physical LUN's + array cache so 100% i/o simply means that a tremendous amount of i/o is going to/from what it sees as one disk. Dd especially if using raw devices will always be faster than cooked io such as cp. Moreover, when you use cp, you are constrained not only by how fast the write is going but also by the speed of the reads.

Your dd's should read should look something like this:

dd if=/dev/zero bs=64k count=1000 of=/xxx/yyy/f1

Writes should look something like this:
dd if=/xxx/yyy/f1 bs=64k count=1000 of=/dev/null

That at least lets you separate the disk related reads from the disk related writes.

You may need to increase your scsi_max_qdepth.