[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: A DNS Based Mapping Peering System for Peering CDNs
The main metrics are time and money. Time might be related to hops or
measured latency; money is the cost of sending the bits. If QOS is
available, then maybe we need to bring in those attributes.
If the tables are exchanged, then there are graph algorithms to
efficiently find a short distance between two points. In the usual OSPF
calculation, each party uses all the other parties tables to find good
paths from its site to all the other destinations.
For the "find the closest surrogate" problem, a "director" must
do this calculation using the client as the origin. It's still the same
calculation, but the storage requirements might be much greater
(because of the multiple origin problem and because "destination"
depends on what object you are trying to access).
>>> "Abbie Barbir" <email@example.com> 11/30/00 08:44AM >>>
Thanks for the reply, please find my notes below.
Hilarie Orman wrote:
> First, there's more than one metric that could be used, so the tables
> must give distance in terms of named metrics.
can we elaborate on those metrics. Can we elaborate on the distances and what do
> Second, in order to do this, CDN's must exchange tables that
> identify their surrogates and measured distances. I thought
> there was some sensitivity about this level of disclosure. Thus,
> the architecture would only require that some form of query/response
> protocol be used; this might be a DNS query.
ok agreed on the sensitivity. However, let us assume that all tables are shared,
the distances that are measured
still do not reflect the distance from the surrogate to the client. If it does
can you explain how??
> Third, shortest-path-first is a reasonable way to minimize a graph
> metric parameter for those parties who are willing to exchange
ok this can be done, but again are we minimizing the graph to get to a node that
is not the client. If not can you elaborate?
> >>> "Abbie Barbir" <firstname.lastname@example.org> 11/29/00 02:50PM >>>
> Martin May wrote:
> > We work on this issue. Indeed, we are about to propose the exchange of
> > "mapping tables" (PMT) during the peering process. A tree based DNS-MS
> > system simplifies the peering exchange especially if footprints are
> > used.
> good to hear that.
> on a different note, any ideas on methods that could choose a particular
> surrogate that is closest to the client. It could be beneficial if this aspect
> of mapping start getting more attention. Any feedback people????