Hi Folks

I have 2 questions regarding the clr profiling api. I am using the gcp_profiler sample as the basis for the questions. Using ver 1.1 of runtime, vc7

Background

I am working on a project with a public web service interface. This interface is called by a client program. We have tracked some performance issues down to being within the web service code. As far as timing a requests, we rely on the timing information collected by the client. So when a client says a request takes 3 seconds, i would like to know what part of that 3 seconds is spent in the suspect components. This info is not usually available in the comercial profilers i have seen, as they summarise the data for many requests into a single record identifying busy components / classes

Questions

1. When hooking the profiling api with a modified version of the gcp_profiler, my program obvoiusly runs far slower, as expected, but is it incorrect to say

Function x entered at 12:20:1.111111111111
Called 200 child function
Leave x at 12:20:1.99999999999

that x actually took .8 to run or am i actually timing a whole bunch of interactions with the profiler that are sqewing my results

2. The callback requested via the SetFunctionIDMapper() method can request that the Enter and Leave methods for a functon not be invoked. My guess is, that this saves time / improves the speed of the profiler. I am basing my decision on wether i want timing information on the "namespace.class" value that the function belongs to. The callback seems to occur twice between JITCompilationStarted and JITCompilationFinished. My problem is, that apparently the function metadata is not available until after JITCompilationFinished, so i cant use that callback to check the name of the "namespace.class" to disable enter/leave callbacks to that function. Is there another way of handling this

Many Regard

Anthony Novak