This discussion has been locked. The information referenced herein may be inaccurate due to age, software updates, or external references.
You can no longer post new replies to this discussion. If you have a similar question you can start a new discussion in this forum.

Is there a best practice for Polling Engine Load Balancing?

FormerMember
FormerMember

Is there a best practice strategy for this? We have a brand new installation of Solarwinds and it's already loaded up after 2 weeks of adding systems. I chose the easy route and split up the polling by technology, It almost works except we have double the Windows servers as we do anything else. Here are our polling engine % utilizations and node counts

Primary Polling Engine - 43% load Network, VMware, and Storage nodes

APE 1 - 83% load Unix/Linux Nodes

APE 2 - 87% load Windows nodes

APE 3 - 0% load Nodes yet.

This is NPM/Virtualization polling only.

For planning purposes, we haven't added SAM application monitoring yet, so I could easily see use having enough app monitors( IIS, Exchange, SQL Server, and other apps ) to add 43% load in the next month. I could assign those to APE 3, but APE 2's load is already bothering me since I have a lot  more Windows servers to add, around 25% more than we have now.

How does every balance their polling ?

For Windows/Linux servers, I do not poll loopbacks interfaces or removable storage devices, but I included network drives. Should I drop network drives?

For switches, I limited polling to uplink interfaces and loopback interfaces, so I think we are good there.

  • A polling engine is a polling engine and it doesn't care if your technologies are split or not.  You just want to make sure none of yours are above 100%, and ideally 85% (to allow for resource growth, etc for example when you add SAM).  If you have all of your Windows Servers though on a single PE and it is at 87% chances are once you add SAM app monitoring you will be far above 100% so sounds like you will need to split up your servers (definitely Windows, and probably Linux) across APE 3.

  • FormerMember
    0 FormerMember in reply to john.ta

    I know all of that, I'm looking for an algorithm or strategy that will keep the load balanced. The first goal is to balance the polling load, a secondary goal is to take remove decision making from the process, so that I can document how to add nodes to Solarwinds without there being any ambiguity in how it's done. Right now, selecting the polling engine is a not a clear-cut process that I can document and hand off to junior tech's to use for adding nodes.

  • The strategy is to fill up APE 1 until that is as full as you would like to make it, and then move on to APE 2 and so on.  It doesn't matter if you have servers spread across different PE's and co-mingling Windows and Linux.  Your junior techs should be adding everything to highest numbered APE, and as soon as APE 3 fills up you create APE 4 and they use that one.

  • Ideally you want to mix your type of load across the polling engines so you can shift load around at will rather than being dependent on one particular servers for any particular load.

    the module engine aka BusinessLayer can be configured to run a separate process for each type of polling; you MIGHT want to take advantage of that if you are in a very large environment. However this adds a layer of complexity that most sites don't need, and is only needed if you're running servers at the limits of their performance, and I would not worry about it until then (as there are other things to be concerned about).

    Our process here is to add all new stuff to the main server, and then about once a month move all of the polling load from that server off onto the least loaded APE. This allows us to check out anything new/odd/unusual with new equipment easily and balance load across the other servers. it also keeps the polling load on the main engine light which improves performance of the webUI and other components that only run there.

    Here are two of my polling engines; note that one of them as a much higher Wireless polling rate than the other (because it happens to have more wireless controllers on it)

    Polling Completion 99.99

    Elements 9542

    Network Node Elements 2198

    Volume Elements 63

    Interface Elements 7281

    Routing Polling Rate 1% of its maximum rate.» Learn more

    UnDP Polling Rate 0% of its maximum rate.» Learn more

    Hardware Health Polling Rate 25% of its maximum rate.» Learn more

    Wireless Polling Rate 1% of its maximum rate.» Learn more

    F5 Polling Rate 0% of its maximum rate.» Learn more

    Polling Rate 42% of its maximum rate.» Learn more

    Total Job Weight 4075

    Elements 8974

    Network Node Elements 2175

    Volume Elements 0

    Interface Elements 6799

    Routing Polling Rate 0% of its maximum rate.

    UnDP Polling Rate 0% of its maximum rate

    Wireless Polling Rate 13% of its maximum rate.» Learn more

    Hardware Health Polling Rate 21% of its maximum rate.» Learn more

    Polling Rate 40% of its maximum rate.» Learn more

    Total Job Weight 4240

  • FormerMember
    0 FormerMember

    It looks like the way to go is to add to the lowest number server until you hit the 85% mark, then load up the next.

    Something I noted above is that I am monitoring network drives because they were selected by default during discover/import process. Is there a way to 'unmonitor' network drives in batches? Since we have thousands of servers, doing it one node at a time would take a very long time to complete.