Question on uniti testing results...

  • Thread starter Thread starter Tom Olson
  • Start date Start date
T

Tom Olson

I have written several tests for a unit and they run fine, but I do notice
extreme differences in performance when I only run one test versus running
every test in the unit.

The class that is being tested has several static methods. When I run all
the tests in this unit the timing (captured through QueryPerformanceCount)
returns in 130-140 ticks. However if I just run this test by itself the
same process takes 8800-8900 ticks.

Has anyone else encountered this?

Tom
 
Tom Olson said:
I have written several tests for a unit and they run fine, but I do notice
extreme differences in performance when I only run one test versus running
every test in the unit.

The class that is being tested has several static methods. When I run all
the tests in this unit the timing (captured through QueryPerformanceCount)
returns in 130-140 ticks. However if I just run this test by itself the
same process takes 8800-8900 ticks.

Has anyone else encountered this?
Click to expand...

Do you mean that just that test takes 130-140 ticks? If so, the rest of
the ticks are probably taken with creating the test fixture, loading
libraries etc.
 
Jon,

The timings of all the test are pretty consistent, however when any of them
are run individually the run time skyrockets. When run as a batch none of
the test show this spike.
 
I have written several tests for a unit and they run fine, but I do notice
extreme differences in performance when I only run one test versus running
every test in the unit.

The class that is being tested has several static methods. When I run all
the tests in this unit the timing (captured through QueryPerformanceCount)
returns in 130-140 ticks. However if I just run this test by itself the
same process takes 8800-8900 ticks.

Has anyone else encountered this?

Tom
Click to expand...

Have you made sure that all methods have been run atleast once
(including the queryperformance methods), before you do the
performance test. Methods aren't JIT compiled until they are accessed
for the first time and JIT compiling takes a good amount of time.
 
Marcus,

I'll give that a try, but wouldn't the timing of at least one of the test be
askew even when run as a batch?
 
Back
Top