Same code poor performance on different server with same RAM and processors

Why same matlab code shows different performance on different server with same RAM and processors?

Below are the server configuration: Code runs 10 times faster on Serve1 than on Server2.

Server1: Windows Server 2008 R2 Enterprise, Intel® Xeon® CPU ES-2697 v2 @2.70GHz 2.70 GHz(8 processors),128GB RAM

Server2: Windows Server 2016 R2 Enterprise, Intel® Xeon® CPU ES-2670 v3 @2.30GHz 2.29 GHz(8 processors),128GB RAM

Answer 1

[Jan]

The processors are different. The E5-2670 v3 has AVX2, but the ES-2697 v2 has not. But this should not result in a factor of 10.

How did you test the performance? Does any other application run on the slower server? Can you identify the bottlenecks of your code and post them?

[Samar Adhikari]

Used profile to check performance. The concerned portion of code (see below) is using matlab built-in function createTask. No other application is running on the slower server(Server2).

        j = createJob(c);
        StartIndex = 1;
        tic
        fprintf(' Task Creation Started \n');
        for count = 1 : NumOfCluster
            EndIndex = StartIndex + NoOfvaluablesperCluster - 1;
            if EndIndex >= size(arrValuableArray,1)
                EndIndex = size(arrValuableArray,1);
                flagtemp = 0;            
            end
                NoOfUsedClusters(counter,1) = NoOfAveraging(1,averaging);
                NoOfUsedClusters(counter,2) = models(modelnumbers);                
                NoOfUsedClusters(counter,3) = count;
          createTask(j, functionhandle, 1, {input parameters...});
            if flagtemp == 0
                break;
            end
            StartIndex = EndIndex+1;
        end
        toc
      fprintf(' Task Creation Finished\n');