- From: Andreas Jungmaier <ajung AT exp-math.uni-essen.de>
- To: Torbjörn Andersson <torbjorn.andersson AT kau.se>
- Cc: discussion AT sctp.de, Thomas Dreibholz <dreibh AT exp-math.uni-essen.de>
- Subject: [SCTP-Discussion] Re: An error in the SCTPlib/SocketAPI
- Date: Fri Oct 31 15:17:01 2003
- List-id: SCTP Discussions <discussion.sctp.de>
- Organization: University Duisburg-Essen
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Torbjörn, all,
please find my comments below:
>
> I tried the setup and compiled your applications. I had them run quite
>
> many
>
> times (~ 100 times) and they never failed, or stalled for me.
>
> Note that this behaviour differs notably from what I had encountered
>
> in version 1.0.0-rc2 (or 1.0.0-rc1 ?) !
>
>
I'm using version 1.0.0 (no -rc*) and socketapi 1.3.1
That's what I meant: I saw the error you mentioned in rc1 and rc2 releases.
It disappeared, however, when I released the sctplib-1.0.0 version.
Are you sure (just to avoid any possible mixup) that all your applications
and socketapilib and sctplib were freshly installed ?
>
> What happens actually is a unidirectional bulk-transfer of data, where
>
> the client application lazily reads data (or rather the respective
>
> thread of the socketapi does). When that happens, the arwnd drops from
>
> about 32K to, essentially, 0. Then, when the application reads the
>
> buffered
>
> data, there is a 200ms "stall", after which an updated SACK (again with a
>
> 32K window) is sent.
>
>
>
> If you can reproduce the problem, I will need an ethereal trace from
>
> your side to help me find out what's going on.
>
>
I have been able two create two tcpdump dumps from a run where
>
everything goes well and one when it stalls. The stall seems to be due
>
to rcwnd which goes down to zero in the runs where it goes wrong.
I suppose, that is advertised rwnd you mean ? Let me investigate that, and I
will report back with my results after the week-end.
>
A secondary problem:
>
Notice also that when everything goes well the client terminates but the
>
server does not retrieve the SCTP_SHUT_COMP notification. The client
>
retrieves its. The server will print its end-state and restart if it
>
retrieves the SCTP_SHUT_COMP.
That may be an issue with the socket-api. Thomas, can you have a look
at this ?
Best regards,
Andreas
- --
Dipl.-Ing. Andreas Jungmaier
Computer Networking Technology Group
University of Duisburg-Essen
http://www.exp-math.uni-essen.de/~ajung
PGP Key-ID: D382 4AC0
PGP Key-Fingerprint: 228C 7C2C 3381 2DD4 9998 2812 5C0B 0B04 D382 4AC0
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
iD8DBQE/onvcXAsLBNOCSsARAiKfAKDjuNBRrQSoV1LApDj++WBr8oFa+wCgppdD
9wbbhFRqapxVRhwz7gZnRMM=
=KUKd
-----END PGP SIGNATURE-----
- [SCTP-Discussion] Re: An error in the SCTPlib/SocketAPI, Andreas Jungmaier, 10/31/2003
Archive powered by MHonArc 2.6.19+.