topic Re: asynchronous I/O in Operating System - HP-UX

asynchronous I/O

yongye_2 — Wed, 03 Aug 2005 03:03:54 GMT

Hi,

Now I want to enable asynchronous I/O on HP-UX 11.11 system when I install the oracle. It seems like asynchronous I/O is applied on the file system. I wonder if asynchronous can apply on raw device? Also I want to know what is the risk of having asynchronous I/O enabled. Can someone give me advice? Thanks.

Yongye

Re: asynchronous I/O

Steve Steel — Wed, 03 Aug 2005 04:06:04 GMT

Hi

google search

hp-ux asynchronous io

get

http://www.google.com/custom?hl=en&lr=&ie=ISO-8859-1&oe=ISO-8859-1&client=pub-5840480241224770&cof=FORID%3A1%3BGL%3A1%3BBGC%3AFFFFFF%3BT%3A%23000000%3BLC%3A%230000ff%3BVLC%3A%23663399%3BALC%3A%230000ff%3BGALT%3A%23008000%3BGFNT%3A%230000ff%3BGIMP%3A%230000ff%3BDIV%3A%23336699%3BLBGC%3A336699%3BAH%3Acenter%3B&q=hp-ux+asynchronous+io

good info

Steve Steel

Re: asynchronous I/O

yongye_2 — Wed, 03 Aug 2005 04:17:02 GMT

Thanks for your prompt response.

I paste the Url you provided into my IE, nothing found by google.

Could you please paste one useful doc Url from your search result? Thanks.

Yongye

Re: asynchronous I/O

Sudeesh — Wed, 03 Aug 2005 04:41:18 GMT

Click on this link and then goto Disk section.

http://docs.hp.com/en/1219/tuningwp.html

If you are using raw disk access, your application will be directly accessing lvol(not through Unix buffer). I belive these parameters are not significant in such conditions.

Hope this helps

Sudeesh

Re: asynchronous I/O

yongye_2 — Wed, 03 Aug 2005 07:18:42 GMT

Hi Sudeesh,

Thanks for your reply. But the doc from http://docs.hp.com/en/1219/tuningwp.html does not answer my question exactly. Thanks.

Yongye

Re: asynchronous I/O

Jean-Luc Oudart — Wed, 03 Aug 2005 07:30:31 GMT

Hi,

to my knowledge asynchronous IO is only supported on raw device on HPUX11i.

cf. attached Oracle/HP White paper

Regards
Jean-Luc

Re: asynchronous I/O

Jdamian — Wed, 03 Aug 2005 07:48:29 GMT

Hi

I don't suggest to set Oracle DB in asynchronous I/O.

We have an intensive Oracle DB that was set in async I/O and had problems (all DB files are raw lvols). Due to a reset of a fibre switch, the timeout to switch from an LVM path to the alternate path was not enough for Oracle; then Oracle reported write I/O errors and set raw devices in read-only mode.

We had to set DB back in synchronous I/O.

Re: asynchronous I/O

yongye_2 — Wed, 03 Aug 2005 09:02:30 GMT

Hi Jean-Luc,

Thanks for the Doc. It demostrates a verey good example on how to make a decision between raw devices database and file system database.
But it gives me no idea about the risk by using raw device with AIO.

Thank you again.

Yongye

Re: asynchronous I/O

yongye_2 — Wed, 03 Aug 2005 09:09:10 GMT

Hi Oscar,

Thanks for your real case.

But I wonder why both HP and Oracle recommend to use AIO?

Yongye

Re: asynchronous I/O

baiju_3 — Wed, 03 Aug 2005 10:15:00 GMT

Hi Yongye,

I know that fs_async parameter value needs to be set to 1 in kernel and /dev/async shud have 666 permissions .

Thanks,
BL

Re: asynchronous I/O

Edgar Canaan — Thu, 04 Aug 2005 11:20:48 GMT

1. Make sure you have the right ownership and permissions for the /dev/async file
2. Touch file /etc/privgroup
3. Type the command # setprivgrp â f /etc/privgroup
4. Include entry in the /etc/privgroup:

CHOWN MLOCK

(eg., sybase CHOWN MLOCK )

5. Again type the command # setprivgrp â f /etc/privgroup.

Re: asynchronous I/O

DCE — Thu, 04 Aug 2005 11:30:29 GMT

Yonge,

AIO is faster than synchronous. Your risk is data loss. If you have problems there is the potential to lose the data that has not yet been written to disk. With todays redundant environments (duplexing, SAN, mirroring) this risk is greatly reduced.