🚀 go-pugleaf

Unfortunately, there are several different communities, each of which
use words in different ways.  I have a reasonable familiarity with the
terminology used by PUP, TCP/IP, XNS, and DECnet.  I am not going to
try to speak for ISO.  The Internet community inherits the traditions
and terminology of PUP, NCP, TCP, and to some extent XNS.  In this
community, the term "gateway" normally means something that forwards
individual packets based on the addresses in the protocol headers.
Thus a gateway must know enough about the protocol to pick the address
out of the header.  However normally it knows only about the lowest
layer.  E.g. it would know only about IP (not TCP) or PUPs (not BSP).
Most protocols have special ancillary protocols that gateways also
must know.  E.g. with IP many gateways handle echo requests, generate
packets telling the source to slow down when the gateway's buffers are
filling, etc.  For IP, PUP, and XNS, most gateways communicate with
other gateways to keep track of network topology (EGP or routed for
TCP/IP, RIP for PUP and XNS).  The term "gateway" is always used for
these objects in the IP community.  I am fairly sure PUP also used the
term "gateway".  XNS seems to use the term "internetwork router".  DEC
has a device called a "DECnet router" that seems to be essentially the
same sort of object.

The term repeater is commonly used to refer to something that operates
at the level of physical packets.  Most typically, it forwards
Ethernet packets.  Such a device does not know anything about the
protocols being carried on it.  The original Ethernet specification
used "repeater" to refer to something that connects two Ethernet
segments to form what is effectively a single Ethernet.  Even
collisions must propagate through it unchanged.  So this is a very
low-level thing.  The Ethernet spec uses the term "remote repeater" to
refer to a pair of half repeaters connected by a fiber optic cable.
This operates at the same level as the other repeaters.

I don't know of any IP or PUP specifications defining anything called
a "bridge".  However this terms seems to be used within the TCP/IP
community to refer to devices that forward at the level of Ethernet
addresses, but are not repeaters.  That is, they are store and forward
devices, but they do their routing on the basis of the Ethernet
address of the packet rather than the IP or other protocol address.
This means that they are independent of the protocol.  They may
exchange routing information, but normally do so using a private
protocol unrelated to the protocols of the packets they are handling.
An example of such a device is DEC's new DECbridge 100, and Applitek's
system that connects Ethernets using a broadband backbone.

I am not sure what people call devices that convert protocols.  Examples
would be DEC's DECnet/SNA gateway (hmmm.... I guess that means they
use the term "gateway").  The XNS spec seems to use the term "gateway"
to refer to devices that convert between XNS and a foreign protocol.

The only serious conflict seems to be over the word "gateway".  This
has been used for so many years within the Internet community that it
is going to cause incredible confusion for anyone to use the same word
with a different meaning.  Generally I try to use an adjective with it
to clarify the sense in which it is being used.  Thus I normally use
the term "IP gateway" to refer to the conventional sort of gateway
used by TCP/IP, since it operates at the IP level.  A term such as
DEC's "DECnet/SNA gateway" is also clear enough to avoid confusion.
However if you use the term gateway without qualification to refer to
something that converts protocols, and if you expect your document to
be read by people whose background is either TCP/IP or 4.2, you are
asking for misunderstanding.