I was reading some Juniper best practices for SNMP management and they mention using row by row polling rather column by column. I'm guessing in Solarwinds that would mean using an interface UNDP poller?
We have abount 50 UNDP node pollers setup for various statistics we want to monitor such as fiber light levels/thresholds, Juniper class of service drops, policing drops, interface errors, etc. We originally had thought to just use node pollers so we wouldn't have to manage adding individual interface pollers as we add new interfaces to monitoring. We have about 12,000 interfaces and 1,200 nodes we monitor across 4 pollers if that helps guage how many tables we are adding with all the custom pollers we've added.
I'm wondering if its worth converting to per interface pollers to cut down on the amount of data Orion is polling? We would of course have to convert our SWQL queries for various graphs and reports and change our processes when we add nodes or interfaces unless we can work out a way to automate applying pollers with the SDK tools.
We had no issues with slowness bringing up pages or graphs until after we recently upgraded (SolarWinds Platform, NPM, NCM: 2022.4.1) and now some custom query graphs that used to work no longer are loading and will time out even after lowering the time frame of stats to load. For example we had a graph that showed policed drops that we had set to show last 24hrs but now this graph won't work unless we back the time frame down to a few hours and then its still flaky if it will load. So I'm wondering if we are just pulling too much and we should parse it down even though it would mean more administrative work when adding things or is there something else we should look into? We've even beefed up resources for the data base and have a standalone web server but still have the loading issues with some of the custom queries that were fine before the upgrade. Then reading this Juniper article made me wonder if its just how we have the custom pollers setup.
---------
www.juniper.net/.../nce-snmp-best-practices.pdf
Changing the Polling Method from Column-by-Column to Row-by-Row
You can configure the network management system to use the row-by-row method for SNMP data polling. It has been proven that the row-by-row and multiple row-by-multiple-row polling methods are more efficient than column-by-column polling. By configuring the network management system to use the row-by-row data polling method, you can ensure that data for only one interface is polled in a request instead of a single request polling data for multiple interfaces, as is the case with column-by-column polling. Row-by-row polling also reduces the risk of requests timing out.
---------
