All:

We're running IDS 10.00.FC6 on AIX 5.2, and have been getting these errors:

KAIO: out of OS resources, errno=12

We've got four such hosts that have been running in full production
since May, and we've encountered this problem perhaps four times since
then. (We started on FC5 and had the problem there, too.)

I've been working with IBM support, and they're suggesting using some
pretty aggressive AIO settings at the kernel level; specifically,
they're recommending increasing the max number of servers per CPU
(currently set to 10) by more than an order of magnitude -- anywhere
from 160 to 640. I'm curious to know what settings others are using,
and if anyone else has encountered this error.

It's worth noting that these hosts run a legacy application that has
literally hundreds of databases, and literally HUNDREDS OF THOUSANDS of
tables and indices.

Thanks,

- Tom Girsch
Hilton Hotels

Thomas,

Same goes for us.
Even on 9.40.FCX with AIX.
Mostly for the instances that have a lot of chunks.
Here is what we have set, from smitty aio, and not getting the log entries
anymore.
We have seen this even with aix 4.2 and aix 4.3.

Change/Show Characteristics of Asynchronous I/O

Type or select values in entry fields.
Press Enter AFTER making all desired changes.

[Entry Fields]
MINIMUM number of servers [400]
#
MAXIMUM number of servers per cpu [800]
#
maximum number of REQUESTS [12288]
#
server PRIORITY [20]
#
STATE to be configured at system restart available
+
State of fast path enable
+



Kernoal






F1=Help F2=Refresh F3=Cancel
F4=List
Esc+5=Reset Esc+6=Command Esc+7=Edit
Esc+8=Image
Esc+9=Shell Esc+0=Exit Enter=Do



"Thomas J.
Girsch"
<tgirsch (AT) NOSPAM (DOT) g To
mail.com> informix-list (AT) iiug (DOT) org
Sent by: cc
informix-list-bou
nces (AT) iiug (DOT) org Subject
IDS 10 on AIX

08/16/2007 10:21
AM







All:

We're running IDS 10.00.FC6 on AIX 5.2, and have been getting these errors:

KAIO: out of OS resources, errno=12

We've got four such hosts that have been running in full production
since May, and we've encountered this problem perhaps four times since
then. (We started on FC5 and had the problem there, too.)

I've been working with IBM support, and they're suggesting using some
pretty aggressive AIO settings at the kernel level; specifically,
they're recommending increasing the max number of servers per CPU
(currently set to 10) by more than an order of magnitude -- anywhere
from 160 to 640. I'm curious to know what settings others are using,
and if anyone else has encountered this error.

It's worth noting that these hosts run a legacy application that has
literally hundreds of databases, and literally HUNDREDS OF THOUSANDS of
tables and indices.

Thanks,

- Tom Girsch
Hilton Hotels
_______________________________________________
Informix-list mailing list
Informix-list (AT) iiug (DOT) org
http://www.iiug.org/mailman/listinfo/informix-list

On 16 Aug, 17:35, kernoal.steph... (AT) autozone (DOT) com wrote:
Make number of requests 12288? That nothing. We use something like
600000 on Linux.

Two Oracle benchmarks I saw on Linux used max number of outstanding
requests as 1 million and 5 million!

David.

OK. The question I have (since we don't have the default value for our AIX
kernel) is how do you test to know that the values are even close to be in
the right ball park? Out side of query testing is there something that the
SA can look at so I can show them it is helping?

Since setting the min to about 50 and the make to 100 we have not gotten the
Error 12 but I am thinking there should be some performance gain to
increasing it to a larger value...

Eric



On 8/16/07, david (AT) smooth1 (DOT) co.uk <david (AT) smooth1 (DOT) co.uk> wrote:

Out on the IBM website, good luck finding it, there is a white paper about
tuning these parameters.

We just increased the values until the messages in the online log went
away.

FYI




"Eric Rowell"
<erowell (AT) gmail (DOT) co
m> To
Sent by: informix-list (AT) iiug (DOT) org
informix-list-bou cc
nces (AT) iiug (DOT) org
Subject
Re: IDS 10 on AIX
08/16/2007 02:04
PM








OK. The question I have (since we don't have the default value for our AIX
kernel) is how do you test to know that the values are even close to be in
the right ball park? Out side of query testing is there something that the
SA can look at so I can show them it is helping?

Since setting the min to about 50 and the make to 100 we have not gotten
the Error 12 but I am thinking there should be some performance gain to
increasing it to a larger value...

Eric



On 8/16/07, david (AT) smooth1 (DOT) co.uk <david (AT) smooth1 (DOT) co.uk> wrote:
On 16 Aug, 17:35, kernoal.steph... (AT) autozone (DOT) com wrote:
Make number of requests 12288? That nothing. We use something like
600000 on Linux.

Two Oracle benchmarks I saw on Linux used max number of outstanding
requests as 1 million and 5 million!

David.

_______________________________________________
Informix-list mailing list
Informix-list (AT) iiug (DOT) org
http://www.iiug.org/mailman/listinfo/informix-list
_______________________________________________
Informix-list mailing list
Informix-list (AT) iiug (DOT) org
http://www.iiug.org/mailman/listinfo/informix-list