High Disk I/O

[Steve ward_3]

How can I go about fixing a High Disk I/O?

The Avg queue length is 0.99

I have identified the disk that has it.

How can I fix it?

[Tim Nelson · Accepted solution]

A typical response would be to spread the data over multiple spindles or faster disk subsystem.

How about some more info.

Why type of disk ?
How is it attached ?
What type of disk configuration ?
What type of application ?
What is the service time as shown by sar -d.

[Rita C Workman]

The first thing I look at...what is being written to that disk.

Let's say it's the disk that supports writing logfiles for Oracle. And you have a very busy Oracle database running. I would not be surprised to see high I/O. So spreading it across multiple disks will help. But you will still see higher than usual I/O simply because of what is hitting it.

Now if on the disk has things like a very busy dbf file(s) and folks are dropping other files out there. Than you need to have your DBA's move those dbf files to mountpoints (with different disk) you tell them to write. Same if folks are dropping data nilly-willy everywhere.
Get control of where the data gets put - and enforce it!

They want performance - you control what gets written to which disk.

Regards,
Rita

....cause it's not just a woman thing...it's a HPUX Admin control things !...

[Bill Hassell]

The answer is that the disk I/O is 100% the responsibility of the running programs. Stop the programs and the problem goes away. A badly written program that read tiny blocks, or a database that lacks enough indexes and searches everything serially is in need of repair -- not something that a system administrator can do.

If you cannot get the code fixed, then change all the disks to use 4Gb fiber, install a SAN switch to configure parallel paths and buy a high end disk array with dozens of GB of RAM. The bad code will still read/write inefficiently but the dramatic increase in speed will make the problem less obvious.

[Steve ward_3]

Thanks all

[Prashanth Waugh]

Hi steve,

think its a difficult to find it which process is using more disk. but u can apply some logic like..
1)first check iostat o/p for a disk which is heavily utilised suppose c0t0d0
2)lvdisplay -v |grep "c0t0d0"
3)u will get the idea like this disk is a member of this lv
4)use #bdf |grep
5)then from lv u will get the idea regarding File system name
6)then check which are the users are using that FS
#fuser -cu
7)you will get the pid
8)then grep that pid like
ps -ef |grep
9)it will show the which proccess is running and owner of that proceess.
=========================================
so doing aboe activity u will get idea

Regards
Prashant

[Prashanth Waugh]

Hi steve,

I think its a difficult to find it which process is using more disk. but u can apply some logic like..
1)first check iostat o/p for a disk which is heavily utilised suppose c0t0d0
2)lvdisplay -v |grep "c0t0d0"
3)u will get the idea like this disk is a member of this lv
4)use #bdf |grep
5)then from lv u will get the idea regarding File system name
6)then check which are the users are using that FS
#fuser -cu
7)you will get the pid
8)then grep that pid like
ps -ef |grep
9)it will show the which proccess is running and owner of that proceess.
=========================================
so doing aboe activity u will get idea

Regards
Prashant