Re: failure detection

[marcelo bagnulo braun <marcelo@it.uc3m.es>]

El 18/08/2005, a las 16:13, Paul Jakma escribió:

> On Thu, 18 Aug 2005, marcelo bagnulo braun wrote:
>
> > so far this is exactly what is contained in Iljitsch draft, Jari's 
> > draft and this is the basic concesous we have so far as i understand 
> > it.
>
> Well, unless I'm reading the wrong draft (the one Iljitsch did the 
> presentation on at IETF?), my understanding of the draft is that it 
> explores the various possible solutions and discusses some the 
> considerations, but does not draw a conclusion.

right, but at least as i read it, there are some basic assumptions 
made, including:

- the mechanism used to explore different paths is to change the source 
and destination address
- the host will explore different paths to find at least one available 
if exists (the maximum number of explorations is hence 2^n but there 
may be not need to explore them all to find a working path)
- unidiractional paths will be supported
- no bgp feed to end hosts


what is being discussed is which is the failure detection mechanism and 
eventually what is is the path exploration mechanism (this is more in 
Jari's draft)

> So yes, I'm rehashing things in that draft, but I'm trying to focus 
> the discussion by discounting some of the possible solutions (or at 
> least relegating them to "implementation discretionary" and 
> concentrate on specifics ;).
>
> Is my understanding correct?
>
> > At this point, what we are discussing is how this monitoring 
> > mechanism should look like
>
> Ok, good.
>
> > and how this ULP feedback should look like
>
> I would discount this from consideration, other than to ensure any 
> mechanism allows an implementation to make use of local positive 
> feedback for optimising things.

well, negative feedback is also in consideration

> > > - You can't mandate a shim6 implementation /must/ send probes
> > >   regularly, to allow other side to use for reachability detection
>
> > not sure what you mean here....
>
> You can't have one side expect to get regular 'pings' in order to 
> determine reachability, otherwise local-positive-feedback to suppress 
> probes/keepalives/pings/whatever-you-call-them would not be possible.
>
> > if you are saying that ULP feedback could result in avoiding shim 
> > probing, then i agree and this is exactly what we have been 
> > cosnidering
>
> Great.
>
> > We have also agreed that failure detection need to be unidirection, 
> > the discussion now is whether the proposed mechanisms are actually 
> > unidirectional or whether they have an underlying bidirectional 
> > assumption
>
> Unidirectional would, I guess, require that all the information needed 
> to setup a shim is available in one message.

not really because the idea is that there may be two unidrectional 
paths (one in each direction) so that address pair used for 
communicating in one direction differs from the one used in the other 
direction

regards, marcelo

> regards,
> --
> Paul Jakma	paul@clubi.ie	paul@jakma.org	Key ID: 64A2FF6A
> Fortune:
> Faith is under the left nipple.
> 		-- Martin Luther