topic Re: Async IO is NOT ?? in Operating System - HP-UX

Async IO is NOT ??

Deep C Pandey — Mon, 05 Feb 2007 23:13:23 GMT

Hi all,

i am using oracle 10g RAC(to nodes) over HP-UX 11i R2(on rp3440/9000 server) using shared RAW storage(HP-EVA HSV100).
is it possiable to use Async IO ??
how can i find that either my machine using Async IO or not ??
if my machine is not using Async IO, how may activate it ??

looking from your kind response.

Thanks in advance.
Deep

Re: Async IO is NOT ??

Yogeeraj_1 — Tue, 06 Feb 2007 03:11:12 GMT

hi Deep,

First of all, see is async io is configured on your system.

see the following thread on how to configure and verify:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1077712

hope this helps!

kind regards
yogeeraj

Re: Async IO is NOT ??

Steven E. Protter — Tue, 06 Feb 2007 03:38:23 GMT

Shalom Deep,

Yes async i/o is possible with your hardware setup.

Some will argue as to whether its true async.

By default in this OS is should be active.

SEP

Re: Async IO is NOT ??

Ralph Grothe — Tue, 06 Feb 2007 03:49:57 GMT

Hi,

asynchronous I/O is, as far as I/O performance is concerned, by far the most advantageous technique.
However, on HP-UX it only seems to work with raw I/O, i.e. writing to raw devices, which at its downside may increase the administrative overhead.
From our Oracle DBA, who has access to the so called Oracle Metalink knowledge base, I was given a document that deals on this topic whose ID was "Metalink No. 139272.1".
I am uncertain about copyright issues why I refrain from attaching the doc to my reply.
Maybe you can find someone at your site who can browse the Oracle Metalinks?
The prevelance of async I/O may be detected by the existence of an async I/O device and Oracle DB writer processes which access it.
Since we are only using cooked I/O with our Oracle DB servers (our clients demand that) but all of our Informix DB servers are using it I can only demonstrate for an Informix box .

e.g.

# ll /dev/async*
crw-rw-rw- 1 bin bin 101 0x000000 Dec 2 2005 /dev/async
crw-rw-rw- 1 bin bin 101 0x000000 Dec 2 2005 /dev/asyncdsk

# ps -fp "$(fuser -u /dev/async 2>/dev/null)"
ps: wrong PID number
UID PID PPID C STIME TTY TIME COMMAND
informix 29952 29951 0 Jan 8 ? 4:52 /opt/inf_99/bin/oninit -v
root 1945 1 0 Dec 28 ? 200:55 /opt/inf_96/bin/oninit -v
root 1947 1946 0 Dec 28 ? 72:37 /opt/inf_96/bin/oninit -v
informix 1812 1 0 Jan 8 ? 219:56 /opt/inf_altdb/bin/oninit -v
root 1961 1 0 Dec 28 ? 44:49 /opt/inf_97/bin/oninit -v
root 1964 1963 0 Dec 28 ? 53:46 /opt/inf_97/bin/oninit -v
informix 29950 1 0 Jan 8 ? 20:33 /opt/inf_99/bin/oninit -v
informix 1818 1813 0 Jan 8 ? 240:09 /opt/inf_altdb/bin/oninit -v

As you can see with Informix the crucial DB processes are called oninit.

You can also check for the existence of the async driver in your running kernel.

# strings /stand/vmunix|grep async
fs_async
max_async_ports
Maximum Number of asyncdsk Ports That Can be Open At One Time
asyncdsk
asyncdsk_included
fs_async 1
asyncdsk_included
asyncdsk

# lsdev -d asyncdsk
Character Block Driver Class
101 -1 asyncdsk pseudo

And last but not least, a couple of async I/O related kernel tunable must be set, like these:

# kmtune -q fs_async -q max_async_ports
Parameter Current Dyn Planned Module Version
===============================================================================
fs_async 1 - 1
max_async_ports 50 - 50

# kmtune|grep aio
aio_listio_max 256 - 256
aio_max_ops 2048 - 2048
aio_physmem_pct 10 - 10
aio_prio_delta_max 20 - 20

Re: Async IO is NOT ??

Deep C Pandey — Tue, 06 Feb 2007 05:03:11 GMT

# strings /stand/vmunix|grep async
asyncdsk
asyncdsk_install
Maximum number of async IO operations that can be specified in lio_list call
Maximum number of async IO operations that can be queued at any time
Greatest delta (slowdown factor) allowed in async IO request priorities
asyncdsk_open_ports
max_async_ports
Maximum number of open asyncdsk ports
fs_async_writes_on
fs_async
async_debug_symbols
# lsdev -d asyncdsk
Character Block Driver Class
101 -1 asyncdsk pseudo
# kmtune -q fs_async -q max_async_ports
kmtune is a wrapper script which exists for compatibility reasons only.
The underlying command used is 'kctune'. New or modified scripts or
procedures should use kctune directly.
About to execute: kctune -q fs_async -q max_async_ports
Tunable Value Expression
fs_async 0 Default
max_async_ports 50 Default
#

Re: Async IO is NOT ??

Ralph Grothe — Tue, 06 Feb 2007 06:13:25 GMT

No, you don't seem to have fs_async support enabled in your kernel.
The mentioned Oracle Metalink describes exactly the steps needed to enable async I/O for Oracle.

HTH