MWA vs vmstat

[John Mannion]

Hi there,

Question on vmstat, and Measureware performance
metrics on HP-UX 11.11 Superdome.

I am seeing a very large discrepancy between vmstat blocked jobs and MWA.

Using both tools to look at performance metrics vmstat on a 5 minute sample shows 'b' column values > 10 over a period of 4 hours.

The equivalent column in MWA GBL_DISK_SUBSYSTEM_QUEUE (I believe) shows values of 0 over the same 4 hour period.

I realise that underneath the hood they are both very different (KI vs pstat), and general conclusion is KI is more accurate ... but I would'nt have expected to see such a huge discrepancy.

Can anyone explain why there would be such a difference ? i.e What have I missed/mis understood ?

Thanks for all/any help

John

[Geoff Wild]

Well...from vmstat man page, it says the b column can be "Blocked for resources (I/O, paging, etc.)"


Whereas the GBL_DISK_SUBSYSTEM_QUEUE is only for the disk I/O subsystem.

As an example, the GBL_DISK_SUBSYSTEM_QUEUE values can be low, while the APP_DISK_SUBSYSTEM_WAIT_PCT values can be high. In this case, there are many processes on the system, but there are only a very small number of processes in the specific application that is being examined and there is a high percentage of those few processes that are blocked on the disk
I/O subsystem.

I got that from:

http://ovweb.external.hp.com/ovnsmdps/pdf/ovpa450_hpux_dict_perf_metrics.txt

Maybe some of that will help.

Rgds...Geoff

[John Mannion]

Thanks Geoff...

Still am unsure though whats causing this
discrepancy though.

Does anyone know what the equivalent
parameter(s) in Measureware are for the
vmstat "b" column ?

Thanks

John

[Steven E. Protter]

Shalom,

Meaurement discrepencies come from how the tools were written and how often the data is collected.

Its also possible there is a bug. Is the system patched up to date?

SEP

[BUPA IS]

Hello John,
As Geoff suggests they are measuring different things, there is no direct mwa equivalent that I know of .
MWA splits things down in to different measures so that you tell indvidual components apart and identify which ones need attention .
the b column in vmstat simply lists all the processes that can't be put on the run queue for an i/o related reason including disk i/o tape i/o paging and I think things like NFS lan and terminal
have a look a the B page in glance to get an idea of the things your system is waiting for.

Mike

[John Mannion]

Hi,

Thanks everyone. I realise both tools are very different underneath the hood. I guess I was just trying to understand why on the face of it the "b" column in vmstat indicated a possible issue but we're not seeing any indication of an issue in MWA data.

I will explore Patch level to make sure nothing was missed here.

Once again, thats for all your help

I'll post anything I find out ..

John