Re: Poor performance related to egp?


Robert Hinden (hinden@ccv.bbn.com)
03 Nov 86 18:07:28 EST (Mon)


Rick,

You raise a number of issues which I would like to respond to.
Firstly, while everyone has been aware for some time that GGP does
not do a good job in regard to exterior EGP routes and there is no
plan to fix GGP, I disagree that it doubles the overall load on the
Arpanet. It certainly does cause some traffic which is forwarded
through a core gateway that is destined to a network behind an
external EGP gateway to get an extra hop. Also, as you point out
that for traffic that originates behind a EGP gateway that is also
destined to an exterior network will get an extra hop. These extra
hops do not double the overall load on the Arpanet. If one were to
count the traffic from all of the cases which do not cause an extra
hop (e.g. host to host, core gateway to core gateway, core gateway to
host, host to core gateway, host to exterior gateway, exterior
gateway to host, etc.) I suspect that over 80% of all Arpanet traffic
could be accounted for.

The problem with the Arpanet, in my opinion, is that it is under
configured for the traffic that is presented to it. Getting rid of
all extra hops will not fix the Arpanet. There needs to be more
capacity (trunks and IMP's).

The extra hops, while not optimal, aren't the worse thing in the
world. I think most people would agree that it was more important to
make the core gateways handle more networks than fix the GGP extra
hop problem. Also note that as the Internet grows, it might change
from the current flat routing model to a hierarchical model. This
might cause all traffic to get an extra hop or two.

The Butterfly Gateway located at CSS, which connects the Arpanet to
Satnet, has been the least reliable of all of the Butterfly Gateways,
sometimes restarting several times a day. We installed new software
last Friday which should fix these problems and it has been up since
then without any restarts.

Bob



This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:36:59 GMT