2 Replies Latest reply on Sep 6, 2011 9:18 AM by Karlo.Zatylny

    Concern with interface availability


      We have a big conern around the interface availability stat's.  I think everyone is going to have an issue with the DB Maintenance no matter how many interfaces they have.  If i have done the math right we are looking at 100 milliion rows per day and with a 15 day detail retention rate 1.5 billion row detail table with our 140k interfaces and 120 sec status polling.  I hope this brings up that we need partitioned tables!!!

      1440 / 2 = 720 * 140,000 = 100,800,000 * 15 = 1,512,000,000

      In production today our interface traffic table has 336 million per day and here is how long it takes. 9 minute stat polling 15 days

      2011-08-24 07:28:28,695 [1] INFO  SolarWinds.Data.DatabaseMaintenance.MaintenanceEngine - Summarizing detailed data from InterfaceTraffic
      2011-08-24 09:17:38,165 [1] INFO  SolarWinds.Data.DatabaseMaintenance.MaintenanceEngine - Summarizing hourly data from InterfaceTraffic
      2011-08-24 10:04:29,650 [1] INFO  SolarWinds.Data.DatabaseMaintenance.MaintenanceEngine - Removing older data from InterfaceTraffic


        • Re: Concern with interface availability

          That's a good question!

          • Re: Concern with interface availability

            To help address this issue we have provided the following features and options which will be available during the RC.

            1. Interface availability will roll up status polls and store a row into the _Details table at a rate of at most as frequent as your statistics collection.  (So by this it will store data every 5 polls given the default poll values - so a _Details record that had 4 up and 1 down poll would store a 0.8 value as a weighted calculated value)  Therefore, by default, the InterfaceAvailability_* tables will be at most as large as your InterfaceTraffic_* and InterfaceErrors_* tables from a row count perspective.

            2. Interface Availability historical storage to the database can be turned off on the entire system (to be have like pre 10.2 versions) or turned off on an interface by interface basis.

            3. The InterfaceAvailability_* tables will have their own summarization schedule so that you can keep data for less (or more) time than your other tables. 

            Please give us any feedback about these items that you would like to.