Differences in results when profiling the same app on a VPC vs. hardware

Question

Would one expect to see differences in results when profiling the same application on a VPC vs. results using Instrumentation?

My coworker and I are profiling the same application.  The application  includes the following:  the main exe - unmanaged code, many many libraries & UI,  written in C++ that calls a dll written in mixed mode (c++/C#) that calls a managed C# assembly.  The last assembly calls another managed assemly written by another team - we will refer to this assembly by the name AssemblyXXX.  We are both using Insturmentation to profile - although I working on a VPC and my coworker is working on hardware.  The question is this - for a key method in AssemblyXXX that is not ours he is getting zero (0) for all of the Application #'s for both time and %; were as my numbers are similar for this method between Elapsed and Application time and %'s.  Why is this?   

Another question - If we do the same set of actions should  we expect to see the same results or does the type of machine (VPC vs hardware) matter?  He is seeing some signigicant differences between  Elapsed Exclusive Time % and Application Exclusive Time % running on hardware but I am not on my VPC.

Answer 1

This is expected. On a virtual machine, the Profiler won't be able to distinguish between "Elapsed Time" and "Application Time" under instrumenation profiling because some CPU performance counters are not virtualized (thus not accessible).

Answer 2

Thank you for your response.  So if on the hardware for a method if  we are seeing Elapsed Time % as 35% and Application time % as 0% - does that mean that 35% of the time was spent in the kernal?  So which number should we use?

Answer 3

Approximately Elapsed Time = Application Time + Kernel Time. For more details, please refer to http://blogs.msdn.com/profiler/archive/2004/08/04/208286.aspx. Thanks.