So far we've found that by removing unnecessary services from the COM+
components (something that is way too hard to do, IMHO), we've gone from
a ten-fold performance degradation (I believe he mentioned something
about 3 seconds in NT/MTS and 27 seconds in W2K/COM+) to a two-fold
degradation (I believe he mentioned something different, 600-some
milliseconds to 1300 milliseconds).

We're still looking at the code, but I believe the next problem is
contention for the locks on cached policy sets. Say it takes tens of
milliseconds to run interceptors, and a second to talk to the database.
If you're servicing dozens of active threads, the odds of any contention
at all at the policy set cache is very low (which makes the caching
introduced to COM+ a performance win). But if you remove the second
during which the code lives in the database, then contention for policy
sets suddenly becomes high, and the cache is a perf loose. But this is
just my theory, we're examining that. And if this is the problem in the
demo version, it is not likely to be the problem in the real world
application.

If you measure the performance of cross-context calls in NT/MTS and in
W2K/COM+ in isolation, you'll find that W2K is not slower.

-----Original Message-----
From: Timothy J. Ewald [mailto:tjewaldDEVELOP.COM]
Sent: Thursday, October 04, 2001 9:45 AM
To: DCOMDISCUSS.MICROSOFT.COM
Subject: Re: FW: COM+ and Free Threading

> What I said was, there is no evidence that the performance degradation

> you're seeing is due to changes in the cost of cross-context calls
> between NT/MTS and W2K/COM+. I am not arguing about the existence of
> the symptoms, just Tim's analysis of their cause.

Everyone agrees that cross-context calls are more expensive than
same-context calls because of the interception. The position I took in
my book was that you should only pay that price when you need to. I did
not address the issue of COM+ interception overhead relative to MTS, nor
have I tested it. The topic has come up on this list before and I have
*hypothesized* that increased overhead in interception might be
responsible for the performance decline Charles and others have seen
moving from MTS to
COM+. I suggested reducing contexts as a possible solution to the
COM+problem
Charles is having. I certainly do not claim that that is the only
possible explanation and since I never saw the source code, there wasn't
much I could do to verify it.

> But this is beside the point.

Yes. Something is making Charles' real world applicatoin run slower
under W2K/COM+. So what is it?

Thanks,
Tim-

----------------------------------------------------------------
Users Guide http://discuss.microsoft.com/archives/mailfaq.asp
contains important info. Save time, search the archives at
http://discuss.microsoft.com/archives/index.html .
To unsubscribe, mailto:DCOM-signoff-requestDISCUSS.MICROSOFT.COM

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]