RE: hard questions: request routing

["Dmitri Krioukov" <dima@krioukov.net>]

I'd like to add that the path vector protocols are the worst known
ones from the loop prevention and convergence standpoints (cf.
"Delayed Internet Routing Convergence" by Craig Labovitz
et al., for example).

On the hand, why can't something related to delay be a good
candidate?

thanks,
--
dima.

> -----Original Message-----
> From: owner-cdn@ops.ietf.org [mailto:owner-cdn@ops.ietf.org]On Behalf Of
> Oliver Spatscheck
> Sent: Friday, March 30, 2001 5:41 PM
> To: Eric Dean
> Cc: cdn@ops.ietf.org
> Subject: RE: hard questions: request routing
>
>
> Eric Dean writes:
>  >
>  > > One concern with this approach would be transient routing
> loops.  Simply
>  > > providing a path vector in the RR protocol does not prevent
> transient routing
>  > > loops. The severity of the problem depends on the dynamic of
> the topology
>  > > changes. Providing similar information on a per request
> basis would prevent
>  > > that problem, however, it is a rather ugly solution.
>  >
>  > By topology changes, you mean with respect to the client
>
> Yes I mean the topology of the RRS with respect to the client.
> For example:
>
> CDN A redirects to CDN B for client C
>
> topology change:
>
> CDN A redirects to CDN D for client C
>
>  >
>  > > Another concern here is that the goal for CDI is slightly
> different from L3
>  > > peering (best effort connectivity). CDNs were invented to
> decrease latency,
>  > > increase throughput and reduce cost. Therefore, forcing
> everybody to use a
>  >
>  > Another important differentiator is that IP networks are
>  > destination-based, while CDN's are client source based.
> Request routing
>  > looks at the source of the requestor making content requests.  I don't
>  > understand why simple path-vectors do not address loop prevention.
>
> Simple path vectors prevent loops in the steady state. I was complaining
> about two issues:
>
> 1. Path vectors do not prevent transient loops which can happen during
>    changes in the value of the chosen metric. This changes  have to
>    be visible to all RRS before we can guarantee that there are
>    no transient loops. Since the updates of the metric will be
>    frequent the transient loops will be frequent.
>
> 2. Using a single metric. I was makeing the point that RRS path length is
>    obviously the wrong metric and that I don't think we know THE
>    single metric we can all use to achieve the goals of CDI.
>
>  >
>  > > At this point I think the most sensible solution might be to restrict
>  > > the number of redirections to at most:
>  > >
>  > > content owner->authoritative CDN->second level CDN
>  >
>  > I'm not sure how such restrictions can be imposed.
>  >
>
> They would be enforce by the SLAs.
>
> Oliver