[Mono-dev] Ldap and Threading

[Mike Glenn]

The one thing I've not heard you mention is whether you are testing on a single
or multiprocessor box?
When I was testing the issues they where much more prevalent on a multiprocessor
box than on a single. Also if you reference back to my post with the test code I
state that the code is the simplest version of the issue I could create. I was
originally trying to reuse connections with a simple pooling queue. When doing
that the exceptions within the receiving thread was very prominent. 

But I don't think I can stress enough that the box that saw all these issues the
most was a dual Xeon /w hyperthreading running Gentoo, 2.6 kernel and nptl. If
you'd like me to test the patch you want to submit on that box and give you the
results I'd be happy to do that for you. 

Mike Glenn

> -----Original Message-----
> From: Konstantin Triger [mailto:kostat at mainsoft.com] 
> Sent: Tuesday, January 31, 2006 5:35 AM
> To: JD Conley; Mike Glenn
> Cc: mono-devel-list at lists.ximian.com
> Subject: RE: [Mono-dev] Ldap and Threading
> 
> Hello,
> 
> I'm absolutely sure that your intentions are good. And I do 
> want to take
> a patch, just be sure that the race is really resolved. I'll 
> work on it
> sometime in the next few days and will reply with the results.
> 
> In the same time, if you can provide some more 
> information/explanations
> it would be very helpful.
> 
> Regards,
> Kosta
> 
> 
> -----Original Message-----
> From: JD Conley [mailto:jd.conley at coversant.net] 
> Sent: Sunday, January 29, 2006 8:24 PM
> To: Konstantin Triger; Mike Glenn
> Cc: mono-devel-list at lists.ximian.com
> Subject: RE: [Mono-dev] Ldap and Threading
> 
> > As I wrote in my analysis of your patches, I ran the application you
> > provided and, unfortunately, did not repro the issue. (see
> >
> http://lists.ximian.com/pipermail/mono-devel-list/2006-January
> /016771.ht
> > ml).
> 
> It didn't work reliably. Now it does. Our intentions are good. Search
> for me on this list. I did some similar (very large) patches to
> Mono.Security that are working quite well. Run your test 
> suite if you're
> worried about introducing bugs into the library. Please, just take the
> patch! I don't want to maintain my own copy of this library.
> 
> Threading bugs are notoriously difficult to reproduce. If I remember
> correctly (Again, I submitted this way back in August -- with only the
> necessities patched), the issue was more easily reproduced on a slower
> network connection where search results trickled in a bit more. The
> background thread that parses the network data was adding results into
> the collection while I was pulling them out from my thread. I was
> testing over a VPN with about 30ms ping latency with the 2.0 build.
> 
> -JD
> 
>