async device driver

[Rammig Claus]

HPUX 11i
Oracle 9i
Async IO on raw devices (Clariion 600)

What is the best minor number for the /dev/async-device?
0x000000 default
0x000001 IR

Kind regards
Claus

[Ken Hubnik_2]

Have you tried 101??

[Ken Hubnik_2]

Sorry wrong number 0x000004

[John Dvorchak]

I am not sure there is a "best" minor number. As long as the major and minor number combination are unique. So in light of that I guess your "best" minor number for the example you gave would be 0x000002

Good luck,
John

[Michael Tully]

To set these up properly.

# mknod /dev/async c 101 4

Mkae sure that you change the owner and group if required. e.g. If your using sybase, then sybase should be the owner.

[Donald Dawson]

I had this same question recently and this is from my research confirmed by HP. We use Informix and hardware mirroring, so we used 101 0

---------------

If using HP hardware/OS redundancy or mirroring, create the async device as follows:
/usr/sbin/mknod /dev/async c 101 0

Otherwise, if using IBM Informix mirroring or no hardware/OS mirroring, or no mirroring at all, create the async device as follows:
/usr/sbin/mknod /dev/async c 101 4

chmod 0666 /dev/async

Donald

[Rammig Claus]

Hi Donald,

I have read somewhere in this forum, that Oracle crashes with a minor number of 0x000004.

Claus

[Dietmar Konermann · Accepted solution]

Hi, Claus!

The 0x000004 bit is indeed an important one. It controls if also retryable IO errors should be passed to the upper layer (i.e. the application).

If you use software mirroring in the application then you need to get the errors conditions passed up... the the application needs to be able to recover from the situation using e.g. its own mirrors. The bit in fact instructs the kernel to pass such errors to the application.

If you use no software mirroring in the application then this flag MUST NOT be set! Otherwise LVM may NOT able to use e.g. PVLinks and mirroring features to hide error conditions from the application!

Best regards...
Dietmar.

[Rammig Claus]

Hi Dietmar,

as we will use Oracle on this filesystems, we must not use the 04 minorcode for /dev/async. Do you know the exact explanation of minorcode 02 "flush the CPU cache after reads".

Kind regards
Claus

[Dietmar Konermann]

Hi, Claus!

I found this info about the 0x2 bit of asyncdsk. Interesting is, that the meaning is just the opposite. :-) The flushing is disabled by setting the bit.

Please note, that only the application vendor can tell for sure, what options should be used with his application! So don't take this comment as a recommendation to set this bit, please.

0x2)
If set, the CPU cache will not be flushed after a read request. This extra cache flush is only needed if the application touches data buffers while they are being read from a device. Setting this bit can improve performance slightly.


Best regards...
Dietmar.