Large differences in time measurements?

freddyg

Hi,

I'm using the profiler for a while now.

I'm profiling a large linq query. When i switch to 'milliseconds display' (F8) i cannot explain the large differences between multiple runs. Recently I installed windows 8.1 and i'm not sure if this has something to do with my issue..

From the last four measures i get the following timings in the profiler: 787ms, 1406ms, 723ms as 1252ms.
My own added logger gave these values (for the complete function): 2435ms, 2498ms, 2456ms and 2519ms

My issues is that the profiler isn't giving equal measurements for the same code on multiple runs. I double checked the hit count, and it is always on 1.
Im following the same steps everytime when i start a profiling session.


How can the difference between the profiler and my logger be explained?
Why is there such a big difference between sessions? (723ms and 1406ms is almost a doubling in timing which i cannot see in my own measure results..)

Is this something i can fix on my side or is this an issue in the profiler software?


Sincerely,
Freddy

Jessica R

Hi Freddy,

Sorry to hear you're seeing some weird results with the profiler!

I'm afraid I'm not immediately sure what''s causing this, but could I please check a few things..

-What profiling mode are you using? Does it make any difference to use a less detailed/more accurate mode such as method-level timings or sampling?

-Do you have the "adjust timings to compensate for overhead added by the profiler" turned on or off?

-Are you looking at CPU or Wall Clock time in the profiler?

freddyg

Hello Jesica,

I'm profiling IIS - ASP.NET on Line-level & method-level timings, all methods inc framework.

'adjust timings to compensate for overhead added by the profiler' is active (this is the default setting i believe)

The timing options is set to CPU time on milliseconds.



I can try a few different settings if you need more information but i'm worried about the inconsistent timings i get..


Thanks,
Freddy

Jessica R

Hi Freddy,

Would it be possible for you to share the application with us so we can try to reproduce the inconsistencies in-house?

Jessica R

Hi Freddy,

So sorry, I've just realized this is for an IIS website which probably won't be easily shared.

Could you kindly try different options (method level timings - only methods with source, disabling the overhead compensation) to see if that gives more consistent results?

It could also be helpful if you could share one or a few of your results files with us--please let me know if this is possible!

freddyg

Hello Jessica,

Sorry, I couldn't respond earlier.

How can I share some files with you? I did 2 tests with method level and line level. I saved the output and also my own 'measurements'.


Sincerely,
Freddy

Jessica R

Hi Freddy,

No worries! We have a support ticket open for you (#15764)--I've just set up some FTP space for you to upload to and sent you the details through the ticket.

Please let me know if you don't receive that email!