Investigation of high %wio

[Sameer_Nirmal]

Hi Guys ,
As a part of system monitoring, I have found high %wio ( max. 85% ) occuring which should be within 20-25%. Now I want to investigate the exact cause of the problem. Firstly I want to isolate the cause from hardware and OS perspective. Secondly I want to know the behaviour of application and database from the system statistics. Following are the system details
rp7410 with HP 11.11 OS / 8GB RAM
SAN ( EVA 5K with 2/16 BS Brocade SAN switches)
Oracle 8.1.7 database
I have collected system statistics and attached herewith. Let me know what you make out from these and your advise or suggestions etc. to improve the situation.
High points to SMART replies.....
Cheers
Sam

[Stefan Farrelly]

Well, your stats are only mildly interesting. Basically your Oracle database is hammering your I/O - asking for data at a much higher rate than your disk subsystem can supply it, causing a high wio%. Thats not basically such a bad thing - your I/O subsystem is working as fast as it can, which is good. Its just that you want it to go faster ?

Firstly you need to explain your I/O subsystem. Its all fibre to an EVA - is it 1Gb fibre to 2Gb ? How many fibre connections do you have to your EVA, 1 or 2 ? or perferably more. Are you running SecurePath - which will balance all I/O across multiple fibre connections - which really helps I/O.

Secondly - how is your EVA configured ? EVA's create volumes across all spindles in the array, which means one server using it could slow down the entire EVA for other servers using it. If you use a lot of heavy I/O on your EVA across multiple servers you should configure your EVA to leave some spindles completely free for certain servers - to isolate them from being affected by what other servers are doing (basically a mini SAN).

Thirdly - you can confirm your EVA I/O performance using an HP tool called stkio.hp - do you have this tool ?
For example, we run it on our servers with EVA's and get over 380MB/s. It would be very interesting to see what result yours gives, but you need to run it when your apps and users are off your server.

[Sameer_Nirmal]

Hi Stefan,
Thanks for very quick reply....

I/O subsystem :- Server is connected to each EVA box with Fibre cable and 2GB HBA. We have 3 fibre channels to each EVA with Secure Path v.30C

EVA Configuration : We have created volume groups with raw logical volumes. Pls advise as how can I check EVA configuration.

I don't have stkio.hp tool. If u have it, pls send to me to confirm EVA performance.

Sam

[Stefan Farrelly]

Hi Sam,

You need to see how many disk groups you have on your EVA and how many disks in each. connect to the EVA and go to Command View EVA then note how many disk groups you have and how many disks in each.

Email me and I can email back the stio tool;
stefan_farrelly@ipcmedia.com

Cheers,

Stefan

[Sameer_Nirmal]

Stefan,
I have two EVAs consisting of two disk groups
EVA01 and EVA02. EVA01 has 9 disks and EVA02 has 8 Disks.


Anyone here who can advise about good configuration of hardware and oracle database which should be used for this setup by considering the attached statistics?

Thanks
Sam

[Stefan Farrelly]

Hi Sam,

ive emailed you the tool, lets see what results it gives.

Cheers,

Stefan

[Sameer_Nirmal]

Stefan , I am waiting to get a downtime from customer to use stio tool.
Guys,
As I understand that if u distribute the disk I/O across multiple disks would balance my Disk I/O load. If you see the attached statistics, disks like c50t0d0 , c51t0d1 are hot disks. Those disks have many dbf files of Oracle. Is it possible to relocate some of dbf files to less used disks?
Anyone had done this before?
Any genious there who ( HP and Oracle Guru)
can advise on the best resolution to my issue?
It would be appretiated...

Cheers
Sam