Why the Interrupt State is so High??

[lunew]

After configed as a Cluster node via Memory Channel,the CPU resources is serious lacking. I find it occupied by Interrupt and MP Synch,but before they are lower.
The monitor cpu mode is as attachment.

Who can give me a clue?

Thanks a lot!

[Arch_Muthiah]

Lunew,

The link below give good explanations about DLM, Interrupt stsate/stack saturation, lock request latencies,... etc. Hope it will be very useful.

http://www2.openvms.org/kparris/bootcamp2005_lckmgr.pdf

also this has been discussed in this forum earlier, you can refer that too.

Archunan

[lunew]

Archunan:

I will try it.

[Hein van den Heuvel]

>> Why the Interrupt State is so High??

Because the application (unknowingly)requests that? What is it doing?

Not trying to be cute here. (well, ok, a little :-)
The system, left alone, would not show this behaviour.
So the application is triggering this.
It is likely to be locking activity.
It may or might not be reasonable, or fixable, but there is no way for us to help you more without knowing a little bit about the applicaition.

Given all the exec time it could be RMS.
Or is it a database?
What exact OpenVMS version? What lockmanager settings?

'Before' was it a single system, or a cluster with a different interconnect?
Did it ever run 'right'?
What seemed a reasonable idle:user:exec:kernel:mpsync:int ratios?
Did the user/process count change recently?

It the clusters usage balanced? Any user, for any applications part on any node? Any chance to focus certain users/data areas on specific nodes (partitioning)?

Any layered products (communications, caching tools, defraggers,..)

Hope this helps a little,
Hein.

[Arch_Muthiah]

Lunew,

I found another usefull link given below, which explains, how to configure mem.channel V2.0 cluster interconnects for
1.Two MEMORY CHANNEL PCI adapters and a single cable
2. Thre or more nodes confign.

http://h71000.www7.hp.com/openvms/memchan/index.html

http://h71000.www7.hp.com/openvms/os/ovmsv71_functionality.html#heading_1.1

Here there are they are talking about the version number of adaptor and cables much match mem.channel hub version.

Please find time to gr thru this, may be usefull?


Archunan

[Andy Bustamante]

Hein asks the golden question. What did this figures look like before the memory channel and clustering?

Other questions. Are you using host based volume shadowing and are any volumes in merge/copy state? Are you using fibre channel or scsi disk?

Andy

[Volker Halle]

Lunew,

performance problems are always difficult to diagnose in a forum like this. You need to be prepared to answer lots of questions and may be target of lots of speculation.

A tool like T4 and it's ability to correlate different sets of collected data would help a lot. Ideally, you would have had it running on the old config and can compare before-and-after scenarios - but that's probably too late ;-(

Try to get an idea of certain aspects of your system load with some MONITOR commands to see whether you can identify the kind of operations causing this massive INT and MPSYNC load:

$ MONI DLOCK
$ MONI MSCP
$ MONI IO

Spinlock tracing would be the last resort to find out, where the CPUs are spending their MPSYNC time.

Volker.