Showing results for 
Search instead for 
Do you mean 

TruCluster LSM : High disk I/O

Advisor

TruCluster LSM : High disk I/O

I have a TruCluster with 2 Alpha servers and 2 BA storage box. LSM is being used.I have 1 disk in each of the storage box forming a disk group with RAID0 (mirroring).They form one volume. The problem i am having is that i am getting high disk I/O . The collect output for these 2 disks shows 99.99% BUSY sometimes.
Do you think if i add 2 more disks and create a new volume under the same disk group , will the disk I/O decrese ?
4 REPLIES
Honored Contributor

Re: TruCluster LSM : High disk I/O

>> .I have 1 disk in each of the storage box forming a disk group with RAID0 (mirroring)

Mirroring is typically called RAID-1
RAID0 is STRIPING.

>> The problem i am having is that i am getting high disk I/O .

And why is that a problem?
That's why you have those disk right?
Perhpas you suspect application slowdown due to this?

>> Do you think if i add 2 more disks and create a new volume under the same disk group , will the disk I/O decrese ?

NO. the total disk IO will not decrease, but the PER DISK disk-io might depending on HOW will you add those drives. If you manage to re-balance the IO over all drives after adding them, then you may be able to improve application throuput and/or response times.

LSM will NOT re-balance on its own.
You need to help.

ADDING MEMORY, may help more to reduce disk IO.

First and formost you need to know how those drives are used.
O.S disks?
Swap?
NFS?
Oracle?

And you may want to figure out the RW ratio.

Try commands like
- df
- swapon -s
- hwmgr -show scsi
- volinfo

Good luck!
Hein.
Advisor

Re: TruCluster LSM : High disk I/O

Thanks for the response.
It's RAID1 and not RAID0 .

The disks are used for Oracle and every second there's lots of select and update queries happening for the Oracle DB.

The dg is now 66% used (in terms of space). Total space being 68G .

According to you LSM will not redistribute the data if i add one more "RAID 1" volume to this disk group??

Will it only start writing on the new disks when the old disk capacity is used up ?

If not LSM , then how can i redistribute the data? I cannot use another RAID level.

Thanks for your suggestions,
Honored Contributor

Re: TruCluster LSM : High disk I/O

>> The disks are used for Oracle and every second there's lots of select and update queries happening for the Oracle DB.

So ask the DBA (you?) where the IO is going, and whether it is a performance concern. Statspack can help a lot with that
To optimally address this you need to understand whether the IO is datafile related, temp, redo, undo,...

>> The dg is now 66% used (in terms of space). Total space being 68G .

Largely irrelevant.

>> According to you LSM will not redistribute the data if i add one more "RAID 1" volume to this disk group??

If you add a plex to the group then you have only added freespace which will not be used. Agree?

Now you have to put it to use.

>> Will it only start writing on the new disks when the old disk capacity is used up ?

Pretty much. Oracle datafile fill only grow on the volume they were created on.

>> If not LSM , then how can i redistribute the data?

By moving files around.

One way to use it is to add a volume onto the new space, and move an oracle datafile or two from the old busy disks to the new disks.

To properly do this you need to know which files to move to obtain some balance. Oracle STATSPACK, or UTLESTAT can be of great value picking the right files to move.

The easier solution, and oracle recommended practice is to 'stripe and mirror everything'

>> I cannot use another RAID level.

Yes you can. LSM will gladly give you extent based striping on top of the mirroring.

Check out ... volassist layout=stripe.

Ideally you backup all data; drop volume, re-create the volume but now striped volume;
and restore all;

You need free space or taped + guts to do this.

You may want to check out some of the Tru64 Best practices documents:


http://h30097.www3.hp.com/docs/best_practices/sys_bps.html

And... you may want to consider professional services to help you if this is a real, commercial, issue.

The (lack of) detail in your questions and answers suggests some help is needed beyond friendly speculation in this forum.

Good luck!
Hein van den Heuvel ( at gmail dot com )
HvdH Performance Consulting
Honored Contributor

Re: TruCluster LSM : High disk I/O

Striping will definetly help you but I would not recommend you to chnage the file sytem layout so easily, it should be propely planned and as of now As said above best ways is to put ur DBA on work as you said its Oracle file system,ask them if number of physical reads etc are high or what.
Is this a very heavy database??
you got to troubleshoot and understand where is the bottleneck.

If you add 2 more disks and distribute you file system , it would definetly help but again if oracle is not properly tuned you may end up in same circle.

BR,
Kapil+
I am in this small bowl, I wane see the real world......
//Add this to "OnDomLoad" event