2 Replies Latest reply on May 22, 2013 2:58 PM by RJKSolutions

    Performance Counters agony

      I'm running data through StreamInsight using AF SDK in a pseudo adapter. I wanted to use the PI Interface for Performance Monitor to capture the Performance Counters that are automatically generated by StreamInsight for the Server, Queries and Input Adapters. 

       

      After reading Microsoft's definition for the Performance Counters (http://technet.microsoft.com/en-us/library/hh290543(v=sql.105).aspx?ppud=4) I figured I could still stream the Counters in to a PI Server to monitor. However, it seems that the unique number is not a "one time unique number" based on the machine it is running on (with some index for each instance) but a new unique number generated each time an instance of StreamInsight is created. This means each time StreamInsight is stopped/started then all the PI Interface for Performance Monitor tags created become orphaned.

       

      Anyone come up with any clever ways to keep the tags connected? I may have to end up having the hosting application start the StreamInsight instance, find the Performance Counter prefix and modify all tags before running...but that is far from ideal.

        • Re: Performance Counters agony

          Hello Rhys,

           

          I've been working with some developers on the following information:

           

          Many products treat counters as transient objects to be used in occasional interactive troubleshooting. This conflicts with our PI Interface for Performance Monitoring (PIPerfMon) which was designed with the expectation that the counter name is persistent. If you intend to stick with PIPerfMon tags, a tag modification routine is likely the best because only solution.

           

          We do not have control over the StreamInsight server and have very limited view into the process(es). You are using the AF SDK to stream data into and out of StreamInsight. At that level you have about as much control as anyone can have.

           

          Event Flow Debugger is for debugging and is taxing on performance. So it might help troubleshooting performance issues.  

           

          The Interface Team is considering rewriting the PI Interface for Performance Monitoring in the relatively near future. Your use case looks like an “interesting” test case.

            • Re: Performance Counters agony

              Thanks Gregor.

               

              We are running a number of queries that collectively use the AF SDK but we want to monitor data volume/availability across all queries, not all the data for the query comes from a PI stream. The idea is to predict degrades in query performance across 100's of installations. Perfmon was going to be a quick win until this issue.

               

              Will let you know where we end up.

               

              New PI Interface for Perfmon is interesting, I saw it recently had an update for the first time in years.