7 Replies Latest reply on Aug 1, 2016 1:52 PM by alano

    Poller capacity

    gibjim01

      Hi there,

       

      Is anybody knows here why Solarwinds recommends to poll no more than 10 000 elements per polling engine, no matter how the machine is built? Can a super server, with plenty of RAM, CPU and disk IO, can poll more that 10 000 elements?

       

      For example, can a single server, with 16Gb RAM, 2 quad core 2GHz Cpu, and 4 15k 300Gb disk RAID 10 run more than 10 000 elements, with a 5 minutes polling rate? If not, why? Is it a hardware limitation or an OS related limitation?

       

      Thanks!

        • Re: Poller capacity
          HemiTruck

          super server has no effect it is a limitation of the polling engine they built

          v10.2 is better than all before it

          prior to v10.2 the limit was 8,000 per polling engine

          now with v10.2 it is 12,000 per polling engine

          if you need to do more just purchase an additional poller and double your capacity.

          but the additional poller will need to reside on a separate server.

          1 of 1 people found this helpful
            • Re: Poller capacity
              gibjim01

              Hi HemiTruck,

               

              Thanks for your answer. 12k elements may be true because I am running NPM10.2 and we have performance problems since we reached that number. However, the server resources are not su busy. We have plenty of RAM, CPU and the disk queues are almost always at 0, with some very quick spikes. Within the next year, we will need to poll more than 30 000 elements and the datacenter guy is concerned about racking 3 more servers for polling, with 3 more for failover engines. Since we have a lack of rack space, he suggest me to buy two super VMWARE servers with a shared storage array of 16 15k disks and  install everything in it.

               

              Since the polling limitation is a software matter, is it a good practice to virtualize the pollers? Can 2 pollers live in the same physical machine, but on two separate VMs if we have plenty of CPU, RAM and disk IO? If not, is there a workaround so we would reduce the number of servers?

               

              Thanks!

                • Re: Poller capacity
                  anthonymahoney

                  The new poller in 10.2 allows for 10-12 thousand elements per poller based on standard polling interval.      You could also try off loading the web server function to orion additonal webserver engine.   http://www.solarwinds.com/products/orion/scalability/web_server.aspx   There also should be no problem with running the pollers on a virtual machines.

                  • Re: Poller capacity
                    netlogix

                    I don't think it's solarwinds engine that is the problem, well not completely.  I just don't think the OS can keep up with the number of network connections.  I know solarwinds started to grab more information in one poll, but it still does multiple connections per node and I think that is just killing the network stack of windows.

                    Doing the pollers in vmware would work, I am pretty sure Solarwinds fully supports it.  They don't like the database in virtual (though if the problem isn't database, they don't care).  I have my poller in vmware and it works great (well it's starting to get bad because I am approaching 10,000 and starting to see the pains, but I have some other processes I am gonna move off first to hopefully lower the number of network connections that are open).

                     

                    I wish they changed SAM to something more like NCM where you can separate the poller from the main, but instead they forced you to combine and slow the main poller with all the cool new stuff they gave us unless you get EOC or additional poller licenses.