[ih] origins of the term "router"

[Jack Haverty]

My recollection is that the name change occurred for two reasons.
Maybe I can help with the "why", although my "when" database has never
been very good....

First, in the earliest days of the IP world (late 70s), a "gateway" was
the box that connected between two networks, creating what Vint termed a
"catenet" (for conCATEnated NETwork).  Somebody (is Ginny Strazisar
around?) correct me if I'm wrong, but I think those earliest gateways
had a very rudimentary algorithm (static tables?) for making the
decision on where to send an outgoing datagram.

Somewhere around 1980 or so, the gateway project was moved from one
group of BBN ("Division 4") to another ("Division 6").   These divisions
had names, but of course I can't remember them...  Roughly speaking,
Division 4 did mostly advanced research, while Division 6 did some
similar research, but also was the home of the "IMP Group", where all of
the Arpanet evolution and operations occurred.   I was the guy in Div 6
who managed the Internet-related contracts.  

With the "gateway guys" just down the hall (literally) from "the Arpanet
guys", the "gateways" started to look more and more like Arpanet IMPs,
in terms of internal functionality and architecture.   In particular,
they evolved features like "fake hosts" (an internal virtual host that
would, for example, respond to "ping" datagrams), and "traps". which
allowed remote boxes to report problems back to the Operations Center
("traps" folded into SNMP later).  This was "technology transfer" at
work and we encouraged it by keeping the groups very close in order to
get more of the techniques for network operation into the Internet
levels.

Eric Rosen was one of "the Arpanet guys", and I got him to spend some
time on the Internet project to help inject Arpanet-think into the
Internet.  Eventually we got some things written down in IENs 184,
187-189.  It was in this period that the IP boxes became more and more
like Arpanet IMPs - i.e., they evolved from "gateways" to "routers" in a
technical sense.

At some point (again, date lost in history) we realized that "gateways"
didn't have to just interconnect between networks.  They could be
connected directly to each other by wires (leased lines), just like the
Arpanet IMPs.  A leased line was just a very, very simplistic type of
network - only two ports (one at either end), no packet headers at all,
etc.   I remember giving a presentation, many times, about how the
Internet looked a lot like the Arpanet - boxes interconnected by lines,
some of which were virtual (paths through other lower-level networks),
and some were physical (leased lines).

With the addition of the Arpanet-style routing algorithm (SPF, but based
on hops rather than delay), the boxes were no longer "gateways".   They
couldn't be "switches" either, since that term was owned by the IMPs.
So they became routers.   I don't know who first used that term, but it
was pretty consistently used to identify the boxes that over the 80s
ringed the Arpanet to handle IP traffic with code from the BBN crew -
Mike Brescia, Bob Hinden, Alan Sheltzer, etc.   This is probably the
root of statements like "CMU node became a router" - I bet that
signifies when the new BBN code was installed and Oct 1983 is certainly
in the timeframe.  We were building and deploying "routers" constantly
during those years as Arpa grew the Internet.

The second reason for the name change was more business-driven.
Computer networks had already existed for quite a few years, and all of
the good names were already taken.  In particular, the term "gateway"
was well known within IBM network communities.   It also had the
reputation as a technology that was expensive, unmanageable, and
unreliable.   We "Internet guys" weren't aware of this.

So, from the perspective of a business trying to win more customers,
"gateways" were not the kind of thing you wanted to have on your list of
technologies when you went to try to get another system engineering
contract, sell a clone of the Arpanet, etc.   I remember one BBN
salesman who got booted out of a big auto manufacturer when the CEO
heard he was trying to sell them "gateways"; from then on they were
called "routers", at least by the BBN crew.  I suspect the same thing
happened to startups (Cisco was a startup back then).

As far as I know, no one every really decided to formally change the
name to "router".  It wasn't a standards issue, and there wasn't any
vote or decision or such.  It just happened over time.   I saw it happen
driven by the two reasons above, and there were probably others as well,
especially as the various marketing departments tried to show how their
box was different and new, and much more capable than all the others.

Hope this helps,
/Jack Haverty
MIT 1966-1978; BBN 1978-1990; Oracle 1990-1998


On Tue, 2006-05-16 at 11:42 +1000, Tim Moors wrote:
> I'm wondering if some of the sages on this mailing list might be able
> to shed some light on the origins of the term "router", in particular
> why it was introduced as a term distinct from "(packet) switch" or
> "gateway"?
> 
> The earliest reference to "router" that I can find in RFCs is in
> RFC753 from March 1979, which discusses message (email) routers,
> e.g. "The Router is responsible for maintaining sufficient topological
> information to determine where to forward any incoming Message-Bag."
> Such "application-level" routing is mentioned in several subsequent
> RFCs.
> 
> The first reference to routers in the context of the "network layer"
> appears in the April 1984 "Gateway SIG Meeting Notes" (RFC 898) in
> which Jon Postel mentions "leaving the normal router kernel function
> in charge of forwarding datagrams." and provides some history of "The
> CMU Gateway" which ?became? a "router" in "Oct 83".  RFC 1001 also
> mentions network layer routers, while RFC 1009 from June 1987 seems to
> be the first to define and discuss technical details of such routers:
> 
> "A router is a switch that receives data transmission units from input
> interfaces and, depending on the addresses in those units, routes them
> to the appropriate output interfaces."  "Interface Message Processors
> (IMPs) are packet-level routers."  "a gateway is an IP-level router"
> 
> 
> Is there, perhaps, a relationship between the use of the term "router" 
> and activity of that juggernaut of routers, Cisco? E.g. the first RFC to
> mention Cisco is RFC985 from May 1986.
> 
> 
> Tim Moors
> http://www.ee.unsw.edu.au/~timm/
> University of New South Wales
> Sydney, NSW, Australia
>  
> 
>