[140571] in North American Network Operators' Group
Re: coprorations using BGP for advertising prefixes in mid-1990s
daemon@ATHENA.MIT.EDU (Jessica Yu)
Fri May 13 17:37:59 2011
Date: Fri, 13 May 2011 14:37:21 -0700 (PDT)
From: Jessica Yu <jyy_99@yahoo.com>
To: Kevin Oberman <oberman@es.net>, Dorn Hetzel <dorn@hetzel.org>
In-Reply-To: <20110512213707.2FF721CC2C@ptavv.es.net>
Cc: "nanog@nanog.org" <nanog@nanog.org>
Reply-To: Jessica Yu <jyy_99@yahoo.com>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
>Does no one remember EGP?=0A=0AYes, I remember EGP every well.=A0 When we =
built the NSFNET T1 backbone in 1987, EGP was the only available routing pr=
otocol for exterior routing.=A0 We deployed it and used EGP to exchange rou=
ting information with the connected regional networks.=A0 Initially, it wor=
ked fine but then when the routing table and traffic grew, we observed that=
every 3 minutes, the network performance got a hit.=A0 After some investig=
ation, we discovered that it was due to the fact that EGP did routing updat=
es every 3 minutes by flooding the whole routing table to the peer and the =
process overwhelmed router processor. At that time, the processor on the ro=
uter did both routing and forwarding.=A0 =0A=0AFortunately, Yakov of IBM, K=
irk from Cisco and we Merit were working on the development and testing of =
BGP, which was intended to replace EGP.=A0 BGP does incremental routing upd=
ates i.e. it sends its peer the delta whenever routing topology changes rat=
her than flooding its peer with the whole routing table every 3 minutes.=A0=
It saved a lot of processing power.=A0 In addition, it reduces routing con=
vergence time since BGP sends its neighbors the updates whenever changes oc=
curs. In the case of EGP, it may take as long as close to 3 minutes after a=
route change before the routing table got updated.=A0 In addition, BGP has=
loop detection capability due to its inclusion of AS path information.=A0 =
These were the technical reasons to replace EGP with BGP at the time.=0A=0A=
We worked with regional network reps and started to convert NSFNET to regio=
nal peers from EGP to BGP in early 1990s. I also created the BGP Deployment=
Work Group at IETF to push the deployment of BGP in the whole Internet.=0A=
=0A=0Acheers!=0A=0A--Jessica=0A=0A=0A=0A________________________________=0A=
From: Kevin Oberman <oberman@es.net>=0ATo: Dorn Hetzel <dorn@hetzel.org>=0A=
Cc: nanog@nanog.org=0ASent: Thursday, May 12, 2011 2:37 PM=0ASubject: Re: c=
oprorations using BGP for advertising prefixes in mid-1990s =0A=0A> Date: T=
hu, 12 May 2011 17:15:17 -0400=0A> From: Dorn Hetzel <dorn@hetzel.org>=0A> =
=0A> >=0A> >=0A> > The actual number would be considerably smaller as there=
were large=0A> > (for some definition of large) block assignments of ASNs =
<~1000 or so=0A> > to various academic networking entities such as NSFNet a=
nd regional=0A> > networks as well as other Federal/Military networking org=
anisations.=0A> >=0A> > -dorian=0A> >=0A> >=0A> Well, for one data point, I=
was issued 3492 around Spring of 1994.=0A> =0A=0ADoes no one remember EGP?=
ASNs are MUCH older than BGP. And we were=0Ausing BGPv3 prior to the exist=
ence of V4. We used BGPv4 back in the days=0Awhen Tony Li would chastise us=
for reporting a bug in a 10 day old Cisco=0Abuild saying that we could not=
expect BGPv4 code over a week old to=0Awork. He felt that we should deploy=
new code daily.=0A=0AThe big push was to have v4 available before the old =
PRDB was frozen by=0AMerit/NSFnet. (And, who remembers the PRDB?)=0A-- =0AR=
. Kevin Oberman, Network Engineer=0AEnergy Sciences Network (ESnet)=0AErnes=
t O. Lawrence Berkeley National Laboratory (Berkeley Lab)=0AE-mail: oberman=
@es.net=A0=A0=A0 =A0=A0=A0 =A0=A0=A0 Phone: +1 510 486-8634=0AKey fingerpri=
nt:059B 2DDF 031C 9BA3 14A4=A0 EADA 927D EBB3 987B 3751
