Is there any general guidance concerning the cost of using custom counters?
My experiments show the overhead to be extremely low, but they may be
distorted by JIT optimization or caching or...

Are there any existing published guidelines on this subject?

TIA.

Just using the perf counters in general will give you a fairly significant
performance hit. When integrate custom perf counters into my app, i always
have a way to turn them on and off, usually with a BooleanSwitch object being
set in the config file.

Under normal production execution your app should not use, write or read,
any perf counters. you should just have a troubleshooting flag you can turn
on and off. That way you only get the perf hit when you explicitly want to
trouble shoot.


"Rick Spiewak" wrote:

That seems to be the general opinion. I decided to do a simple test,
comparing the time of a loop with and without incrementing two counters (one
absolute, one a rate counter). Over 10,000,000 iterations, the time
difference was 225 microseconds! While this is almost certainly an artifact
due to caching of the loop, it still suggests that the actuall overhead of
Windows performance counters used via .NET is minimal. It almost looks like
the impact should approximate that of doing the boolean test to decide
whether to use the counter. The fact that Windows itself contains large
numbers of counters seems to indicate that there is little impact.

Do you have an example which shows otherwise?

"john conwell" wrote: