Dear Syon,
It is great that you bring this issue up. Yes. There are multiple
systems using location hierarchy to achieve preferential (localized)
peer selection. Two examples I know of are Kontiki (traceroute trees)
and P4P Location Algorithms (two level hierarchy).
During the revision process, we discussed multiple times the possibility
of introducing hierarchical location. In the current draft
(http://tools.ietf.org/html/draft-penno-alto-protocol-03), we added a
sentence (at the top of page 28):
Note that it is possible that a tracker may use only
the Network Map to implement hierarchical peer selection by
preferring peers within the same PID and ISP.
Clearly, this topic can benefit from a "high-bw-low-delay" discussion
setting during this IETF meeting.
Best,
Richard
sy ding wrote:
Hi all:
I am wondering if we need hierarchical PID structure to facilitate
P2P tracker to organize its peers. As I know in some current P2P
applications, they organize the peers into a hierarchical structure
according to its location.
Best
China Telecom Guangzhou Institute
Syon Ding
2009/7/16 Richard Alimi <[email protected]>:
Hi Antonio,
Thank you very much for reading the draft and your feedback. See below:
I found a new version of draft-penno-alto-protocol
(draft-penno-alto-protocol-03.txt) and I have a question about it.
Since an ALTO Server presents to ALTO clients its "my--Internet View",
it may be reasonable for this view to be accurate in describing "near"
Network Locations and being less and less accurate when we move away
from them. So my question is: is there a way to represent "the rest of
the world"?
It seems, reading section 6.3.2.2, that every endpoint is mapped to a
PID (since there's nothing like a 404 "No PID for the endpoint" response
available), so we may group all the "far" networks we have few
information about ("the rest of the world") within a PID. The problem
with this approach would be how to answer to a Reverse PID Property
Lookup on such a PID (enumerating all network maps of far networks?).
Have you already addressed this issue?
When mapping from IP addresses into PIDs, an ALTO Client can use longest-
prefix matching. This will be stated explicitly in the next revision of the
document.
Thus, the "rest of the world" can be represented simply as the 0/0 IP prefix.
The prefix can be assigned to either an existing PID (along with other
prefixes), or allocated to its own PID. This is indeed what we have done in
the P4P tests.
For example, the ALTO Server's reply from Reverse PID Property Lookup can
including the following mappings:
PID1: 128.36.0.0/16 130.132.0.0/16
...
PIDDEFAULT: 0.0.0.0/0
--
Richard Alimi
Department of Computer Science
Yale University
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
