Re: failure detection

[Paul Jakma <paul@clubi.ie>]

On Sat, 20 Aug 2005, marcelo bagnulo braun wrote:

> ok, i guess we have come to key point here.

Maybe.

Note again, nothing in shim6 should preclude what you want. I just 
don't think the 'perfect' case you desire should be mandated as the 
default case.

> We agree that the proposed mechanism proposed for the shim is what 
> is needed to deal with all failure modes and to identify if there 
> is at least one working path right?

Yes.

> We seem to disagree about if the cost that implies is worth it, right?

Yes.

> you seem to consider that there are simpler methods that would deal with a 
> significant amount of the common failure modes, in particular the one you 
> detail above.

Yep.

> I guess that probably RFC3178 already provides a reasonable 
> solution that provides a the protection level that you ask for. I 
> mean RFC3178 protects from failures in the edges in a transparent 
> fashion

That's one solution.

Like I said, there are already /lots/ of other ways and as yet 
unknown future ways to cope with routing failures. Ignoring all the 
other possibilities and just mandating n^2 probing in shim6 seems 
unwise to me - particularly when those other ways could be far more 
efficient. (And particularly when, imho, the failure modes you're 
concerned about are not /that/ common imho).

> > A host could have the following default route:
> >
> > default via ISP1-gateway
> > 	via ISP2-gateway

> what if there is a single router in a link of the multihomed site? 
> i mean, you cannot assume that in all links of the multihoemd site 
> there will be as many routers as ISPs the site is multihomed too, 
> right?

That's possible. But I don't know of many ISPs who 'share' their 
routers. And if they do, they must co-ordinate its configuration. Eg, 
if a router serves both ISP A and B, it obviously will not be 
filtering out packets with source of either A or B ;).

> In this point, i guess you end up requiring source address based 
> routing in the multihomed site, in order to allow the end host to 
> force routing through the selected exit ISP and the shim using the 
> source address to actually select the exit ISP hence the shim 
> selecting the source address, i guess

Maybe, depends.

> > ISP1-gateway device X src ISP1-PA-address
> > ISP2-gateway device X src ISP2-PA-address

> not sure what you mean.. are you thinking in something like GSE here?

Yes, as a subset of shim6. Ie, using an 8+8 static mapping for the 
local ULID(s), and using whatever shim6 control messages are needed 
to map the remote ULID to the correct remote locator(s). It would be 
assumed that the ULIDs are composed of a prefix and a host ID, 
obviously. The mapping would only change the prefix.

I think it would be very useful to allow such a mapping, and hence 
allow split/proxy shim6.

> i agree it would be useful but i still not sure how do you deal 
> with security stuff in this case...

There are no security implications to static mapping. It never 
changes. No more than there are security implications to a non-shim6 
host forwarding packets according to a static routing table of 
destination->nexthop.

regards,
--
Paul Jakma	paul@clubi.ie	paul@jakma.org	Key ID: 64A2FF6A
Fortune:
Those who can, do; those who can't, write.
Those who can't write work for the Bell Labs Record.