async io

[Rammig Claus]

Hi,

i have a system with HPUX11i, Oracle 8.1.7 and invoced async io.
With lsof, i saw that the device /dev/async is used.
But in glance and with tusc i saw no aio_reads or aio_writes, only normal reads and writes.
Why.


Let the sun shine

Claus

[Arturo Perez del Galleg]

To control this feature it is a kernel parameter fs_async with value 1 for asynchronous I/O. See in sam.
HTH

[Dietmar Konermann · Accepted solution]

Hi!

First, a Happy New Year to all of you! :-)

Second, the fs_async kernel tunable has nothing to do with the asyncdsk kernel driver. Claus, you should know that already.

Third, it's quite normal that you don't see any aio_*() calls... these are part of the "POSIX asynchronous I/O facility" and have also nothing to do with asyncdsk.

Best regards...
Dietmar.

[Rammig Claus]

Hi Dietmar,

so we cannot differ between "normal" IO's and async IO's? Is it all the same system-call (read and write)?


Gru?? nach Ratingen and a happy new year
Claus

[Johannes Seippel]

Hi,

1st of all, the fs_async kernel paramter is intended only for hfs file systems. JFS does not support fs_async. So it is not of much use unless you are using HFS filesystems.


Back to the Oracle - /dev/async question:

To enable the asyncio enhancement that defers memory locking to improve application startup times, create the
async device file with minor number 256 using the following commands. The user must have super-user rights
to execute these commands.

Delete the old device file.
# rmsf -v /dev/async

Create the new device file
# mknod /dev/async c 101 256

Note: This minor number should only be used on systems that have enough physical memory so that paging is avoided. Paging can cause serious performance degradation with this new enhancement. On systems where paging is an issue, this minor number should not be used.

The effective user ID of the process calling async driver, typically called by a process for database applications suchas Oracle and Sybase, must be a superuser or the user must be a member of a group that has the MLOCK privilege.

To check the privilege capabilities for a group, issue the command:

/usr/bin/getprivgrp

If the output of getprivgrp(1) does not indicate that the group has the MLOCK privilege, it can be set by issuing the following command as root:

/usr/bin/setprivgrp MLOCK

[T G Manikandan]

Do you have any parameter like
DISK_ASYNC_IO =true in your init.ora file .

in Oracle
Revert

[Rammig Claus]

Hi Johannes,

First: Happy new year.

Second: I only know the minor number 0,1,2,4,5,7 of /dev/async.
What is the minor number 256?


Gr????e
Claus

[Dietmar Konermann]

Hi, again!

Currently I don't have any idea how to exactly track asyncdsk IOs. I will see if I can find something more.

The bit 8 (adding 256 to the minor) is a new one... for deferred memory locking.

This is from JAGad74567 (PHKL_24457 and later):

A new device minor number 256 is provided. When the driver is opened
with this minor number it ignores the request to lock the SGA and
instead locks the buffers when an IO is started (in the performance
path) and unlocks the same when the IO completes.

Note: This minor number should only be used on systems that have enough
physical memory so that paging is avoided. Paging can cause serious
performance degradation with this new enhancement. On systems where
paging is an issue, the minor number 0 should be used.

Best regards...
Dietmar.

[Rammig Claus]

Hi Dietmar,

can you tell me the difference between the
POSIX asynchronous I/O facility" and the asyncdsk.

Thanks in advance ...
Claus

[Dietmar Konermann]

Hi, Claus!

The asyncdsk implements is a propriatary interface, desgned especially for DB vendors to access raw devices.

For the POSIX async I/O facility see the aio(5) man page. This is an open POSIX standard... also for regular files.

Best regards...
Dietmar.

[Rammig Claus]

Hi Dietmar,

if this is so, then the aio_physmem_pct kernelparameter has nothing to do with /dev/async.

When using /dev/async, is there a way to change this parameter in the same manner?

Best regards ...
Claus

[Dietmar Konermann]

Hi, Claus!

No, this is not configurable from the kernel side. It's the application's task to allocate the memory. All you can configure is max_async_ports.

Ciao...
Dietmar.