Hi Sebastian,
On Tue, Aug 10, 2010 at 1:23 AM, Sebastian Kiesel <[email protected]> wrote:
> Rich, all,
>
> This is a comment about section 9.6.3. Protocol Extension:
> 'Location-Only' Peer Selection in draft-ietf-alto-protocol-05:
>
> I think this proposed extension is problematic, as we would have to
> define what an ISP is, and "same ISP is good" is not true in all
> scenarios.
Point taken. The first concern is something that Jon mentioned during
the meeting and the second concern is something that was mentioned in
the draft as a consideration (basically, ALTO Client would need to
double-check in the Cost Map).
> Furthermore, I think we don't need additional protocol mechanisms
> for implementing your idea: Just use the "relative operator's
> preference" rating criterion (see ALTO reqs, ARv05-16). Configure
> the ALTO server to answer with "1" for IP addresses / PIDs inside
> the ISP and with "0" otherwise. Technically, this is the same
> as your approach, but it does not introduce new mechanisms and it
> avoids having to define what an ISP is.
This is certainly one way of implementing it. However, to avoid
suggesting that the ISP enable costs for a particular peer selection
algorithm, it may be easier to suggest that the ALTO Client try to
estimate which PIDs are within the ISP and which are outside. (Of
course, if the ISP really does wish to expose only very coarse-grained
information like this, that is fine.)
How about this for new text at the end of the section:
This algorithm makes two assumptions about the preferences
communicated by the Network Map:
o The ISP prefers peers within the same PID to peer with each other
(see Section 4); and
o The ALTO Client can distinguish between peers within the same ISP
and peers outside of the ISP. In implementation at the ALTO
Client, it may may estimate a threshold based on costs read from
the Cost Map.
What do you think?
Thank you very much for the comments,
Rich
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
