SCTP Discussions

Text archives Help


Re: [SCTP-Discussion] helping Multi-Homing work faster


Chronological Thread 
  • From: dp AT ii.net
  • To: Andreas Jungmaier <jungmaier AT web.de>
  • Cc: discussion AT sctp.de
  • Subject: Re: [SCTP-Discussion] helping Multi-Homing work faster
  • Date: Tue Feb 21 03:42:02 2006
  • List-id: SCTP Discussions <discussion.sctp.de>

Hi Andreas,

Thanks very much for the reply. I can see now that keeping the whole
recovery within SCTP is going to be the best scheme. I'll wind back the
default timers and see how it recovers. It was also useful to see the
reference to 800ms requirement for a message path to recover for SS7.

For those how haven't seen the paper, it compares an association with 2
paths, against 2 separate associations. The first scheme allows for a
better average performance on failure, and is likely to be a better scheme
in most situations.

cheers,
davidp.
--
David Pascoe,
mailto:dp AT ii.net,
Western Australia


On Wednesday, 15 February 2006, at 11:26:37 [GMT +0100] you wrote:
AJ> Hi Dave,
AJ> you may want to have a look at an old paper of ours, which you can find
AJ> on
AJ>
http://tdrwww.exp-math.uni-essen.de/inhalt/forschung/sctp_fb/sctp_links.html
AJ> It's named "On the Use of SCTP in Failover-Scenarios", and you
AJ> probably find some insight there.

AJ> Kind regards,
AJ> Andreas

AJ> David Pascoe
<dp AT ii.net>
schrieb am 15.02.06 08:32:32:

AJ> Hi All,

AJ> I can see that you get multiple paths when you create an association with
AJ> a multi-homed peer. Assuming that both ends are multi-homed, what could an
AJ> ULP do to speed up the multi-homing feature ?

AJ> I assume a heartBeat will bring down a path and choose another path when a
AJ> fault occurs.

AJ> Is there anything that the ULP do to help a path swap over. If say a local
AJ> interface goes down, we could guess that a path within the association
AJ> _might_ be gone because we would route out the down interface to the peer
AJ> address in that path - but ... it isn't a given that they route _to_us_
AJ> over the same interface that we route to them.

AJ> So I was wondering what people do to give positive event indications to
AJ> sctplib rather than wait for heartBeat to do it for you ? (also assuming
AJ> that you aren't running a routing protocol that will hide all of this from
AJ> you).

AJ> cheers,
AJ> davidp.
AJ> --
AJ> David Pascoe,
mailto:dp AT ii.net,
Western Australia







Archive powered by MHonArc 2.6.19+.

Top of page