How to improve Disk I/O ?

[Fragon]

L2000/HP-UX11.00
18G HD x 2 (Mirror)
DB size : 1.5G
========================
My DB size is increasing quickly. And the system speed become very slow. I use:
#sar -u 10 10
I found the %wio is 95% or more. Of course this is disk bottleneck.
Adding more disks is impossible because of the budget. And, for data safety, the Mirror can't be broken.
How can I improve my disk's performance though tuning OS's or File System's parameters? or How to do?
Any suggestion is appreciated!

[Bill Hassell · Accepted solution]

The disks are doing what they are told to do by your database program. It is common to want the kernel to magically improve the performance, but very little improvement can be made with kernel changes for your system.

You will get a huge improvement in speed by tuning the database program. The goal is to reduce the number of I/O's to disk and this can only be done with significant changes to the database parameters. Note: you may need a *LOT* more memory to take advantage of database tuning.

[Michael Tully]

Do you have on-line JFS? If so it is possible to tune some areas other than utilising 'delaylog' in conjunction with the 'mincache' option.

e.g.

/dev/apps/myfs /myfs vxfs rw,suid,delaylog,datainlog,mincache=tmpcache 0 2

Now, looking at your hardware and what your actually doing, you may not get any improvement at all. You would need to really spend some money on a disk array that can handle the volume. The reason is that most of the performance gain is in the cache that the disk arrays use and not utilising your actual server hardware.

[Patrick Chim]

Hi,

Does the OS reside on this two disks also, i.e. all the OS, application and DB are running on these two disks only ? If so, I am afraid there is no other alternative to improve the IO performance unless you use a faster disk or buy more disks to share the IO !

Does your system got any other disk array such as SC10, AutoRaid or do you got a SAN over your network ?

Regards,
Patrick

[Michael Tully]

scrub the option with mincache... It won't work. It should only be used for temporary filesystems.

/dev/apps/myfs /myfs vxfs rw,suid,nolog 0 2

[Fragon]

Hi Bill, thank you very much for your answer.
You said that tuning DB's program , I agree too. But, a very simple program such as delete each records of a table over 60,0000:
for each tbl_name:
delete tbl_name.
end.
I found when running this program, the %wio is over 95% too. How can I tuning this program? (Obvious Impossible.)
Would you give any other tuning way to improve this?

Thanks in adv.

[harry d brown jr]

Actually you've already answered your own question by stating that there isn't money in your budget to replace or improve your disk sub-system. If the Company doesn't realize that data is money then it's time someone told them what's up, and if they wish to remain in business they have to pony up to the bar and pay their tab.

live free or die
harry

[Jean-Luc Oudart]

1) I think that the OS and application/database should be separated. Therefore, 2 disks (mirrored) aren't enough

2) You say that one simple program is to empty a table.
Are you using Oracle ?
If yes I suggest you use the truncate table command, this is immediate

my 2 cents

Jean-Luc