[En-Nut-Discussion] Heave load will harm the system?

[Henrik Maier]

Harald,

I have a modded version of the httpserv.c example where I observe the 
sockets end up in the CLOSEWAIT state after frequent connecting and 
reconnecting (typically caused by repeatedly pressing ctrl-R in the 
browser window for a few seconds). Once one manages to get all available 
4 sockets into CLOSEDWAIT state, the system refuses all further 
connections and never recovers.

However my changes to httpserv.care not minor either and it could relate 
to the use of Adam Dunkel's protothreads for connection multiplexing 
rather than using the Nut/OS threads. On the other hand the TCP state 
machine should be robust enough not to stall in a transitional state 
like CLOSEWAIT.

I might relate to the observed issues which are discussed here or not, I 
haven't investigated it further due to other priorities.

I am happy to pass on the httpd_protothreads version for further 
investigations.

Henrik
http://www.proconx.com

Harald Kipp wrote:
> At 21:17 28.02.2006 +0000, you wrote:
> 
>> i believe last time the system hang cause by a suddent memory drop, 
>> and not able to recover from that drop,
>> however when i redo the test, it seem everything just fine.
> 
> After several reports with systems hanging, I'm a little
> bit alarmed. Let's see what we have:
> 
> 1. As far as I can say, most users experienced problems with HTTP.
>    Several use the original httpd sample with "minor" changes.
> 
> 2. A few reported problems without HTTP. May be they do a lot
>    of TCP connect/disconnect cycles?
> 
> 3. Some users reported problems with large networks. Unfortunately
>    no simple test application is available.
> 
> 4. Several very complex applications are running without problems
>    for months. I assume, that most of them use long lasting
>    connect times (compared to HTTP).
> 
> 5. No one reported low memory situations and Michael's tests
>    with ARP flooding additionally proofed, that the APR cache
>    doesn't seem to cause any problem.
> 
> 6. Some reported problems after upgrading from 3.4 or earlier.
>    Again the applications are complex and the problem may be
>    caused by different stack usage.
> 
> Please correct me, if I'm wrong.
> 
> Harald
> 
> _______________________________________________
> En-Nut-Discussion mailing list
> En-Nut-Discussion at egnite.de
> http://www.egnite.de/mailman/listinfo.cgi/en-nut-discussion
> 
> 
>