3 Replies Latest reply on Jan 8, 2015 1:45 PM by david.novak

    Efficiency of DistributedCallable in Ispn7

    david.novak

      Hi,

      I'm using Ispn 7.0.2 with the distribution mode together with persisent data in file-store (config attached). Ispn is embeded into my system; I need to execute some "operation" on a set of keys and I am using DistributedExecutorService and DistributedCallable in order to forward my "operation" to all nodes in the system, let them read a set of keys from their local storage and process the operation. I am testing the efficiency on a variable number of objects read from the cache and with variable query frequency.

       

      When I am using one Ispn node, it works well but when I run Ispn in distributed mode on two nodes (data is split half and half, machines in the same rack), the overall performance of this operation is WORSE than in case of one node... It seems like if the distributed processing (des.submitEverywhere() ) was sequential instead of parallel. When I was using Ispn6 (config attached),  there was a speedup when I was adding nodes (as expected) but it is vice versa with Ispn7.

       

      Isn't there any configuration of the thread pools that is actually sequential in Ispn7 by default? For instance, I found setting <transport executor>, but I don't have a clue where to find possible executors or if it could change anything... Or any other ideas, please? I'm desperate...

        • 1. Re: Efficiency of DistributedCallable in Ispn7
          pruivo

          Hi David,

           

          how are you iterating over the keys in the DistributedCallable?

           

          Cheers,

          Pedro

          • 2. Re: Efficiency of DistributedCallable in Ispn7
            vblagojevic

            David,

             

            Any updates on the issue you encountered? It seems there are many moving parts and variables in your application and potential cause of the slowdown could be hidden in any one of these. If you want to isolate and focus on distributed execution I suggest that you try PiApproximationDemo from Infinispan examples. In your network setup try a few runs of PiApproximationDemo with a single node and a few runs with multiple nodes and recored/observe results.

             

            Regards,

            Vladimir

            • 3. Re: Efficiency of DistributedCallable in Ispn7
              david.novak

              Pedro and Vladimir,

               

              Thank you for your replies. I am stuck with some other work at the moment and I will get to the issue within a few weeks... I will certainly try the PiApproximationDemo and will inspect other variables in depth.

               

              Best,

               

              David