Re: minimizing the cost of managed/unmanaged collaboration

> > Okay, so I have a tentative plan to convert significant
> amounts of C++
> > into managed code,
>
> Sorry, I don't want to be annoying, but since you are writing
> that you plan to convert significant amounts of code - do you
> have a performance improvement plan that _needs_ this
> conversion to be done?
I have an objective, and general metrics, but not a plan.

> I know that you currently have COM interfaces for the
> interop, but converting them to C++ Interop or IJW (VC++
> 2005/2003) doesn't necessarily solve your performance
> problems, because your parameters still need to be marshaled
> with every cross-managed/unmanaged call.

Actually what I'm planning will eliminate some of the crossover, so there will be 1/2 the
data marshalling and many fewer marshalled calls.

> And even if COM is your culprit, converting your code to
> unmanaged DLLs and consume them from C# via P/Invoke might
> require much less work, although only you can estimate this.

Understood.

> Therefore - before rewriting everything, you might want to
> look at the monitoring stuff in the article I mentioned,
> maybe you can find out the actual reason for your performance
> problems. Or, if these don't help you at all, at least try
> out this rewrite on some small portion of code to measure the
> benefits. Unless you are really sure that rewriting will pay off.

Good advice.  I didn't find too much helpful for monitoring.  I retrieved that CLR Spy but
it seems to just report bad interop activities; I don't see any performance indicators.  I
will definitely try simple changes first before diving headlong into massive changes.

Alan

===================================
This list is hosted by DevelopMentor®  http://www.develop.com

View archives and manage your subscription(s) at http://discuss.develop.com