CPU Utilitization messages on only 1 node and cpu

Biite · **Joined:** Thu Sep 01, 2011 8:27 am **Posts:** 20

Hi,

We've setup threshold alerts via SSMC, one of them is for the CPU utilization.
We've set the following parameters (output of 'showsralertcrit'):

Code:

System_CPU_Utilization_Major    enabled hires  major    cpu  -recur=10/10,idle_pct<10

Now we get major alerts like the one below, but only on 1 of our 9 array's:

Code:

System Reporter threshold alert System_CPU_Utilization_Major with condition idle_pct<10,recur>=10/10 has been satisfied by cpu on node 0:1 with value(s) 0.0,10 recurrences

Two things are strange:

The error always shows on node 0 and cpu 1
Look like there's no averaging happening of CPU usage

Anyone know what process(es) are running on node 0, cpu 1?
Anyone know a better way to alert on CPU usage?

Regards,
Martien

Biite · **Joined:** Thu Sep 01, 2011 8:27 am **Posts:** 20

Logged a case with HPE for this issue. Now wait to find out what the problem is

ailean · **Joined:** Wed Nov 09, 2011 12:01 pm **Posts:** 392

Maybe check shownode to see which is master and shownet to see which is the Active management port?

Any management tools/plugins talking to it other then SSMC?

Is it a 2 node doing lots of stuff?

Maybe graph CPU/Disk/Host IO for correlating spikes during the CPU busy?

MammaGutt · **Posted:** Wed Sep 06, 2017 7:55 pm

2-node system running dedupe and seeing high cpu util on the non-master node?

Biite · **Joined:** Thu Sep 01, 2011 8:27 am **Posts:** 20

Update...

Dedupe defragmentation process always runs on this node and core, which is a fairly CPU intensive task, but only for about 3 to 4 minutes.

My threshold is set for 10 occurrences in 10 hi-res samples with a cpu idle percentage of 10. So every 5-minutes the CPU idle is less than 10% for 10 times in a row.

That was unexpected for the HPE engineer, he's now checking performance logs etc.

Still waiting

MammaGutt · **Posted:** Fri Sep 08, 2017 3:35 am

Biite wrote:

Update...

Dedupe defragmentation process always runs on this node and core, which is a fairly CPU intensive task, but only for about 3 to 4 minutes.

To my knowledge it runs on all non-master nodes and is a low priority task. So if you had 4 nodes it would run on 3 nodes.

JohnMH · **Joined:** Wed Nov 19, 2014 5:14 am **Posts:** 505

I'm informed this is normal behavior if running dedupe.

Garbage collection runs on any none master node, load distribution will depend on CPG / DDS numbers. If the CPU is busy it will reclaim at a low priority, however if there's plenty of free CPU cycles it will take as much as is needed until it either completes or CPU front end demand increases, at which point it will throttle back again.

So in a two node system running dedupe, one of the nodes will periodically have higher CPU utilization but this is by design.

Biite · **Joined:** Thu Sep 01, 2011 8:27 am **Posts:** 20

It's indeed normal behaviour. In our case the defrag of one of our DDS stores runs for about 50 to 70 minutes, which is just inside the window of the threshold alert defined.

HPE Storage Users Group

CPU Utilitization messages on only 1 node and cpu

Who is online