Operating System - HP-UX
1748004 Members
4544 Online
108757 Solutions
New Discussion юеВ

AIO thresholds how to monitor?

 
Thomas C. McGivern
Occasional Contributor

AIO thresholds how to monitor?

We have an application that is using AIO under HPUX 11.0.
We are having horrible stability issues, our AIO kernel parms are all at defults.
How can I tell just how close to AIO_MAX_OPS or AIO_ASYNC_PORTS AIO_PHYSMEM_PCT is to the threshold values? The vendor is saying "ah, just kick them up 50%" I don't like that approach, I want to know what is happening in the system.
I have mesaureware/glance/perfview, but don't see where those parms in there, don't know if that metric is tracked.

Any help would be appreciated...
Thanks.
5 REPLIES 5
James R. Ferguson
Acclaimed Contributor

Re: AIO thresholds how to monitor?

Thomas:

Document #KBAN00000163 has lots of information on async I/O. Perhaps this will help you.

...JRF...
wyan lowe
Frequent Advisor

Re: AIO thresholds how to monitor?

Hi,

don't know the answer...but

if you have perfview, found out from HP that
"repinfo -s hostname -mwa | grep -i key-word"
will help you find what metrics are available and their correct syntax

"repinfo -s hostA -mwa | grep -i tbl" will show all the metrics available from various data sources that have a "tbl" in its name. This is on hpux 10.20, measureware c.2.10

the later versions (c.2.40) have more metrics available
version info is via "perfstat -v"

I did a "repinfo -s hostA -mwa | grep -i raw" - not sure if those metrics correspond with Asynchronous I/O...(/dev/rdsk/cxtxdx)

Thomas C. McGivern
Occasional Contributor

Re: AIO thresholds how to monitor?

Well.. repinfo returns 20,800 different metrics! but none of them seem to be AIO related (unless the name doesn't contain AIO). Thanks for that info though..

wyan lowe
Frequent Advisor

Re: AIO thresholds how to monitor?

that doc that james mentions looks pretty good
i've attached it anyways

the repinfo | grep -i will sort of work if you know what keyword to search on...
I tried searching on "raw" but don't know if it relates.

here's another idea: you can setup "applications" for glance to watch
via the /var/opt/perf/parm file
you specify application = label
file = *.dbf or something like that.. there are other keys to use besides "file"
if it's an application you're monitoring, maybe this is one approach..
it appears to group the related processes on the same page/display with CPU info & disk info...



If the application you're using can supply the metric info, then you could use "dsi data integration" to make that data available to measureware/perfview and alarmgen...which is cool. there's a whitepaper on it. www.openview.hp.com/library/ click on papers link - there's a good one about glance in advisor only mode...
Thomas C. McGivern
Occasional Contributor

Re: AIO thresholds how to monitor?

Thanks.. that's not a bad paper, but it's just a summary of what's in the books. Which I'd already read, which drove the question, "So.. how much of what we already have allocated, are we actually using?" And their seems no way to easily tell.

Informix is the application, there's some indication "errno=11" that we should probably increase aio_max_ops, but only because we're getting errors do I see that, there's no way within HPUX (that I've found), to be proactive, and adjust upward when needed.

Anyway, thanks for your help, and I'll look into your other ideas in your second post.