Routing WG meeting minutes
Tuesday, 30 January, 2001, Joint Techs Workshop, Honolulu.
WG Chair: ken lindahl <lindahl@ack.berkeley.edu>
Minutes written by ken lindahl, 8 May 2001, from notes taken by Ron Roberts.
Please send any corrections/additions to <lindahl@ack.berkeley.edu>.
----------------
Opening comments
----------------
ken announced that new Engineering Area web pages would be available 14 February
2001, including revamped pages for the Routing WG. WG members are encouraged to
check these out and send any comments to ken or to the WG list. The URLs are:
http://www.internet2.edu/engineering/
http://www.internet2.edu/routing/
[8 may 2001 note: the Chair has gotten no feedback to date from WG members.]
Guy Almes was contacted by Bill Manning and Steven Kent (BBN) regarding the
possibility of Secure BGP testing by Internet2 members. WG meeting attendees
agreed that this would not an official WG activity, but that ken should reply to
Bill and Steven indicating that WG members might be interested in participating
as indviduals.
[8 May 2001 note: I did send mail to Bill and Steven; it was forwarded to
another person in BBN who did not reply. I interpreted this as a lack of
interest and did not press the matter. Any members truly desiring to participate
this are welcome to send me email.]
----------------------------------
I2 vs commodity Internet asymmetry
----------------------------------
ken observed that there is a fair amount of confusion concerning this topic: a
number of folks have commented that asymmetry is the norm in the Internet and
asked why this is an interesting topic. The issue is not asymmetry in commodity
Internet traffic, but rather asymmetry between I2 and commodity Internet, I.e. a
site that is connected to an Internet2 backbone probably expects to both send
and receive data to/from other Internet2 sites via an I2 backbone. If this is
not the case -- i.e. if traffic in either direction is routed via commodity
Internet -- it seems likely that high performance applications suffer. There are
implications for the E2E performance initiative.
The WG charter calls out a milestone of three "case studies" of asymmetry, to
have been completed by this meeting:
http://www.internet2.edu/routing/wg-routing-asymmetry.shtml
This milestone was not accomplished: no volunteers have stepped forward to
perform the case studies, and no sites have volunteered to be the subject of a
case study. Dave Farmer asserted that I2 vs commodity asymmetry is more likely
at sites where there are separate routers providing I2 and commodity
connectivity.
Steve Corbato asked how many folks are running traceroute servers at their
campuses. A small number (~5) responded positively. Bill Cerveny noted that
Surveyor machines will do traceroutes. ken observed that a single traceroute
server provides a view of routing from a single point on a site's topology and
might not reveal asymmetry affecting only part of the site's topology. In
particular, Surveyors are by design located topologically close to sites' I2
connections, where it might be expected that asymmetry is less likely.
Dave Farmer volunteered to help with the I2 vs commodity asymmetry activities.
[8 may 2001 note: I have been negligent by not following up with Dave.]
----
I2db
----
ken presented a small number of slides about the I2db and it's use, showing
examples of objects describing UC Berkeley. A question developed concerning the
use of "NIC handles" in some objects: Mark Prior stated that use of NIC handles
has been deprecated and that a person's first and last names can be used
instead. However, this seems to imply the use of a person-object to provide
appropriate contact information; the person-object has "nic-hdl" being a
required attribute. catch-22. ken promised to follow up with Merit.
[8 may 2001 note: the Merit technical folk agreed with Mark that NIO handles
have been deprecated, but did not provide a full answer to the question about
the required "nic-hdl" attribute. However, I have found an apparently acceptable
alernative: simply use the person's first- and last- names in in the "nic-hdl"
attribute. This seems somewhat self-referential, but seems to work. See slide
NN.]
There was a question about whether the I2db is still at Merit or has moved to
NextHop? [Susan Harris subsequently confirmed that the machine is still at
Merit.]
ken reported encountering problems trying to delete obsolete objects in the
Cable & Wireless routing registry.
ken proposed the name be changed from "I2db" to "I2RR." This is merely a change
to how the routing registry is referred; it does not change any of the names or
values in the registry itself. There was concensus (or at least lack of
objection) among among the WG members present.
Two I2db "early adopters" described their experiences with I2db: Brent Sweeny
(Indiana U and the Abilene NOC) spoke about the beginning efforts at Indiana and
recommitted to get the Abilene policies recorded as well. Cas D'Angelo (Georgia
Tech) spoke about his efforts at getting Georgia Tech into the I2RR. Cas was
tripped up by not knowing what value to use for the "source" attribute in I2db
objects. The correct value to use is "I2".
ken asked if any of those present were intereted in becoming "early adopters";
there were no takers.
------------
Other issues
------------
Dave Farmer reported a problem in which some campus's (not U Minnesota,
apparently) announcements to Abilene were accidentally leaked somewhere in
France by one of Abilene's international peer networks, resulting in undesirable
asymmetric routing. Dave asked whether we should be recommending that sites that
have to announce more specific prefixes to Abilene should also announce the
specifics to other peers? Mark Prior asserted when there is policy that requires
announcing specifics then that policy should be applied globally, i.e. specifics
should be announced everywhere. Matt Davy suggested that this should be an
infrequent proble but that it is worth writing up the tradeoffs of globally
announcing the more specific routes.
This led to discussion of the large number of aggregatable specifics "polluting"
the Abilene routing table. Several folks expressed the opinion that it would be
worth identifying the source and the reason, fixing it if possible, and
documenting the case. ken will solicit volunteers on the mailing list for the
route table investigation. [8 may 2001 note: the Chair has been negligent in
this regard as well...]
