[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-sctp-dev
Subject:    [sctp-dev] (no subject)
From:       chris.hegarty () oracle ! com (Chris Hegarty)
Date:       2011-11-18 14:53:17
Message-ID: 4EC6715D.7050001 () oracle ! com
[Download RAW message or body]

If you are on Linux this would probably be best answered by 
lksctp-developers at lists.sourceforge.net

Sorry I can't be of more help here.

-Chris.

On 11/15/11 03:04 PM, Jens Carlberg wrote:
> Hi again!
> Now that we have managed to run SCTP between different servers, we have
> continued troubleshooting our original problem - bad SCTP throughput on
> localhost.
> - Scenario 1: using loopback interface
> We connect a server sending messages over SCTP to a client. Both run on
> the same machine, connecting over localhost. The traffic starts OK, the
> messages are sent one by one and usually SACKed after two of them, but
> after a while the server starts to bundle the messages and the client
> starts sending a SACK only once every 200 ms. We get a throughput on
> about 185 messages per second. We have assumed the 200 ms is the
> standard SCTP SACK timeout, and by setting it to 1 we have been able to
> see about 14000 messages/s.
> - Scenario 2: using standard network interface
> We connect a server sending messages over SCTP to a client. The server
> and the client runs on two different machines, connecting over a gigabit
> network card. The traffic starts OK, the messages are sent one by one
> and usually SACKed after two of them, after a while the server starts to
> bundle the messages but in this case the client is OK with this and
> keeps sending SACKs in a timely manner. We get a throughput around 50000
> messages/s.
> Once more we are at a loss to explain this behaviour.
> * What differences is there between a real network and the loopback
> interface?
> * Can the SACK timeout be set to 0 for better performance, or a fraction
> like 0.1?
> * We have a dump of the traffic with a SACK for a TSN, that there is no
> corresponding DATA chunk. Is the dump missing data, is there a bug in
> the clinet SCTP stack, or is there scenarios where this is OK?
> //Best regards,
>
> line
>
> *JENS CARLBERG **
> Software Designer*
>
>
> Ericsson AB
> FU Radio System I&V, LI/EAB/FJZ/TE
> Datalinjen 4
> SE-58112 Link?ping, Sweden
> Phone +46 107114141
> SMS/MMS +46 761497941
> jens.carlberg at ericsson.com
> www.ericsson.com
>
>
>
> Ericsson
>
>
>
>
> This Communication is Confidential. We only send and receive email on
> the basis of the terms set out at www.ericsson.com/email_disclaimer
> <http://www.ericsson.com/email_disclaimer>
>

[prev in list] [next in list] [prev in thread] [next in thread]