Understanding polling internals

schwarzkopf

Hi:

I'm looking for informations about poller engines, how does it work.

My Orion NPM haves about 11000 elements and I'm  having chart gaps, but my Orion server doesn'uses total cpu capacity.

We are creating a second poller now, but I'd like understanding how polling works.

Thanks again.

kweise

schwarzkopf,

Polling isn't so much CPU or even memory intensive.  However, it is very disk read/write intensive for the SQL server.  11,000 elements on a single polling engine is a little over the recommended limit (I've generally seen it recommended you have somewhere between 8,000 and 10,000 elements per polling engine.)  If you're running SQL on that server as well, you're almost guaranteed to have gaps in your data. 

I'm sure someone from Solarwinds could give you a much more detailed explanation of how polling works than I can, but I can give you a high level overview of it.  Orion uses both ICMP and SNMP for it's "polling."  The ICMP polls are sent out every 120 seconds by default to monitor up/down status of all of the nodes in inventory.  The SNMP polling is used to collect statistical data (CPU, memory, interface bandwidth utilization, etc.)  for each element in inventory.  The SNMP polling is every 8-10 minutes by default (I can't remember the exact default off the top of my head.)  If your server's disks can't write data to the database fast enough to keep up with all of the polls, you'll see gaps in your charts because the polling cycle didn't complete for some of the nodes in your inventory. 

If you haven't already looked at the following document, it's worth the time to work through.  It gives you some more background on how the polling works and what can cause gaps in your data.  It will also walk you through some troubleshooting steps to help you identify where the bottleneck might be in your environment.

www.solarwinds.com/.../gaps.htm

Hope this helps!