we have a problem with false alarms, which are generated from SNMP Polls. Mainly management agent lost Alarms. We used the Performance Data through the Grafana interface to see that we get maxed out poll threads.
Release : 21.2
Component : Spectrum Core / SpectroSERVER
In busy environments where poll threads are getting occasionally maxed out, the threads can be increased to cope with the load but the caveat to increasing polling threads is that it will most likely cause an increase in cpu as well as notification and Timer latencies.
We can see from the screenshot of the thread usage, there were notification and timer threads maxed as well, all at the same time. Ideally you would increase all 3. You just need to make sure the cpu and total work threads can handle the values. Total work threads may need to be increased, but again at the cost of cpu.