[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [IMRG] directing the discussion
Loki,
> the AS_PATH of the advertisement v the
> path that data takes) - probably the one that interests me most but is
also the
> most abstruse (I am aware of potaroo and the sterling work of Philip
Smith)
Hi. On the forwarding vs. signaling path, you might be interested in the
papers
http://www.cs.princeton.edu/~jrex/papers/sigcomm03.pdf
http://www.cs.princeton.edu/~jrex/papers/infocom04.pdf
See also the slides at
http://www.cs.princeton.edu/~jrex/talks/astrace04.ppt
These papers focus mostly on the difficulty of accurately determining the AS
path the packets actually take, but they also discuss cases where the
forwarding and signaling paths legitimately differ (e.g., due to route
aggregation, packet deflections, and bogus ASes in the BGP AS path).
Actually, our intent when we started this work was to focus soley on
analyzing the cases where they differ, but we found so many mismatches
between the traceroute and BGP data when we applied a prefix-to-AS mapping
(gleaned from the origin ASes in BGP routing tables) to the IP addresses in
the traceroute paths that we had to step back and consider the problem of
how to construct a more accurate prefix-to-AS mapping to use. The biggest
problem was caused by ASes that do not advertise the address blocks they use
to number their equipment, causing the addresses to be wrongly mapped to
whatever upstream provider is advertising the supernet containing these
addresses (e.g., see slide #12 in the powerpoint talk above).
-- Jen
-----Original Message-----
From: imrg-bounces at ietf.org [mailto:imrg-bounces at ietf.org] On Behalf Of Tom
Petch
Sent: Thursday, March 17, 2005 6:16 AM
To: Loki Jorgenson
Cc: imrg at ietf.org
Subject: Re: [IMRG] directing the discussion
I see IETF WG as customers, who could take better decisions if better data
were
available.
In no particular order
routing; there seems very little analysis of BGP and how it is used, what
gets
advertised how (thinking particularly of what does not get advertised and
how,
eg by aggregation or by communities and the AS_PATH of the advertisement v
the
path that data takes) - probably the one that interests me most but is also
the
most abstruse (I am aware of potaroo and the sterling work of Philip Smith)
traffic by application; one frequent discussion point with spam/UBE is the
notion that operators do not care because this is such a small part of the
load,
BitTorrent and such like being the majority use of the Internet; true or
false?
TCP(M); much discussion about the resetting of sessions, what that costs
(slow
start etc), would it be better to behave differently in the event of
unexpected
happenings (RST, ICMP); how frequent are these events?
segment sizes; crops up in IPv6 and network management; when 576 is too
small,
how many networks lose out if the effective minimum PDU needs to be
something
bigger?
congestion; until the advent of wireless, 'everyone knew' that packet loss
was
due to congestion and not to CRC checks; I have looked periodically for
papers
to support this and do not see them; and is wireless quite different, if so
by
how much?
how bit is the Internet in terms of traffic, how fast is it growing? there
was
a question on this on the main IETF list some months ago, I never saw a
reply.
not rocket science, mostly
Tom Petch
----- Original Message -----
From: "Loki Jorgenson" <ljorgenson at apparentnetworks.com>
To: "Jennifer Rexford" <jrex at CS.Princeton.EDU>
Cc: <imrg at ietf.org>
Sent: Wednesday, March 16, 2005 11:57 PM
Subject: RE: [IMRG] directing the discussion
All well received.
And I would add to your list of "who needs to the information" -
applications people or people involved in applications research. These
people are neither experimenting on the Internet or running the
Internet. But they are involved in work that has a direct dependency on
the Internet and networks in general. And they need to characterize the
beast for their own purposes.
This is where I find the "rubber hits the road" and defines relevant
metrics. And I imagine it is where some very interesting
cross-pollination can take place (networking and.... something).
-----Original Message-----
From: Jennifer Rexford [mailto:jrex at CS.Princeton.EDU]
Sent: Wednesday, March 16, 2005 2:52 PM
To: Loki Jorgenson
Cc: imrg at ietf.org
Subject: RE: [IMRG] directing the discussion
-----8< --- snip -----------
In terms of the important question of "who needs the information," I
think
we have at least two camps -- those using data to characterize the beast
(e.g., research folks trying to do experimental science on the Internet,
including work on understanding the performance and limitations of our
existing protocols and/or evaluate new designs) and those using data to
run
the network (e.g., operators trying to detect anomalies, do traffic
engineering, thwart attacks, etc.), though increasingly the lines are
blurry
as more research folks jump in to creating and evaluating techniques for
using measurement data to help operators.
-- Jen
_______________________________________________
IMRG mailing list
IMRG at ietf.org
https://www1.ietf.org/mailman/listinfo/imrg
_______________________________________________
IMRG mailing list
IMRG at ietf.org
https://www1.ietf.org/mailman/listinfo/imrg
_______________________________________________
IMRG mailing list
IMRG at ietf.org
https://www1.ietf.org/mailman/listinfo/imrg