Direct Io and async Io

[hrishy]

Hi

Anybody using direct io and asynco on normal filesystems on redhat advanced server 3 in rpoduction .

any surprises here ?


regards
Hrishy

[marist89]

I'm using async I/O. It's a little tricky to setup, but no problems once it's working correctly.

[hrishy]

Hi Jeff

Thank you very much.Any idea why in mixed environments we neeed to set the parameters like this

filesystemio_options setall

and

disk_asynch_io to true.

How do i find the block size for my ext3 filesystems .

regards
Hrishy

[jlakhani]

Wait a sec.. Isnt Direct IO or ASYNC I/O bad because you are turning off caching either at the file system level or the Kernel Level? If your Unix server caching while Oracle was writing data the FSCK maybe not be able to correct corruption on the effected filesystem since "caching" is turned off?

[pando]

Marist are you sure aio works in red hat? What sort of system calls do you see when you trace dbwr :-?

One Oracle support guy (excellent DBA btw)told me that AIO has not worked so far in Red Hat Linux

[hrishy]

Hi Pando

Have a look at this note on metalink Note:279069.1

regards
Hrishy

[marist89]

Originally posted by pando
Marist are you sure aio works in red hat? What sort of system calls do you see when you trace dbwr :-?

It's been a while since I set it up, but I'm pretty sure it was working correctly. I had to go through lots of changes to make it work, but I was pretty confident it was working when it was done.

[KenEwald]

I know of someone who has it working on RH AS 2.1. His report was that it sped up his i/o big time.

It may work on RHEL 3.0, but I've been working with support for a couple weeks on a TAR and have so far been unsuccessful to "alter database mount" without it hanging.

It's doing something.

output with async io disabled.

$ cat /proc/slabinfo | grep kio
kioctx 0 0 128 0 0 1 : 252 126
kiocb 0 0 96 0 0 1 : 252 126
kiobuf 0 0 64 0 0 1 : 252 126

output with async io enabled.

$ cat /proc/slabinfo | grep kio
kioctx 270 270 128 9 9 1 : 252 126
kiocb 66080 66080 96 1652 1652 1 : 252 126
kiobuf 236 236 64 4 4 1 : 252 126

-Ken

[KenEwald]

Follow-up from my last note...

Applied patch 3208258 and it works now. It would be nice if they'd just put that note somewhere a bit more obvious.

Benchmark test: 1,000,000 row delete

w/o async
2 min 2 sec
iowaits 25-30%

w/ async
1 min 10 sec
iowaits 1-5%

That's a big difference.

-Ken

[hrishy]

Hi Ken

Thank you very much that was really helpful.

regards
Hrishy