DataPower

Hi Suraj,
Performance testing a DataPower workload is typically done one service at a time and then you build a weighted model using a spread sheet to combine them.  So long as the combined set of services does not push above 80% resource utilization the individual services should add linearly, theory of superposition holds. 

If you are focused solely on DP, you will want to design a testcase that ramps concurrency while holding average backend latency constant.  You do not want a static value of latency, you want some variation but stable avevrage over the range of concurrency being used.  You will calibrate the max concurrency to be just more than enough to saturate resource utilization.  From the load driver you will want to collect periodic average latency and throughput statistics.  On the DP you will want to collect dpmon logs, these have all sorts of resource utilization measurement data as provided by the OS, and also poll some of the status providers like HTTPMeanTransactionTime2 and HTTPTransactions2. 

With the load driver's latency and throughput data you will create a graph, latency on Y throughput on X.  This will generate a curve similar to this

The concurrency ramp up is generally ~300 secs worth of total time and its stats summaries are typically ~10 secs.  The dpmon, which is nmon running on DP, samples are 1 sec and those logs are found in default domain temporary://dpmon/ directory.  There are 9 files of 900 seconds each file, ~2.25 hour rolling history.  Rename those files with .nmon extenstion and load them into a tool like https://nmonvisualizer.github.io/nmonvisualizer/ to create time series charts of resource utilization