At 3:30 PM +0200 10/21/13, Stephane Bortzmeyer wrote:
On Mon, Oct 21, 2013 at 09:08:43AM -0400,
Dave Crocker <dhc(_at_)dcrocker(_dot_)net> wrote
a message of 27 lines which said:
He invented the datagram...
Some people say it is Leonard Kleinrock. I welcome any precise
references on the history of the datagram, specially if they are
No, Kleinrock claims to have invented packet switching, not the
datagram. There was an article before his death by Donald Davies,
who looked quite closely at the claims that Kleinrock invented packet
switching as well as Baran and Davies. I have read Kleinrock's
thesis which is the source of the claim and I don't see it. Davies'
very fair article did not see it either.
accessible online (whuich is rare for scientific papers of this
time). Anyway, the fact that someone did something great X years ago
(he also wrote the first shell and built Cyclades) does not mean he
cannot troll today.
I have contended that whether or not you see packet switching as a
breakthrough depends largely on how old you are? ;-) People of the
period from a more telecom background see the idea of breaking data
into small pieces (remember message switching systems already
existed) as a radical idea. If you were a little younger and your
background is in computing, then packet switching in obvious. From a
computer person's point of view, if the problem is posed of
communicating between computers, the solutions is clear: The data is
in a buffer, pick it up and send it. What else would you do? ;-)
Pouzin's insight was that the packets should be routed independently
and not all follow the same path. Furthermore, Louis' reasoned that
no matter how good the network was, the hosts would check to make
sure that everything got there. Hence, he reasoned the network did
not have to do a perfect job, but only make a "best effort" and that
an end-to-end transport protocol would ensure reliability. You
won't find any of this in Kleinrock's thesis. All of this was
implemented in CYCLADES in 1972 (and is well documented in many
papers which are on-line and in subsequent histories). The original
ARPANET did not work this way, although Type 3 messages (datagrams)
were added later. (I have always found it interesting that every
project taken on by Baran and Larry Roberts after 1970 were all
connection-oriented, none were based on the datagram concept.) The
real revolution here is that Pouzin took communications from a
deterministic model of connection-oriented to a stochastic model of
datagrams.
For me, this shift in thinking comes in two stages that I liken to
what we saw in geology: Some people came up with a new idea, packet
switching (continental drift) and after looking at it, it was refined
to the datagram concept (plate tectonics). While from a distance we
often see these changes as sudden and complete in one big insight
they seldom are.
Pouzin is one of the best thinkers I know. You may not like what he
says, but even now he is far from being a troll. His hit rate
exceeds most of us! ;-)
I think his slides do miss some essential points, but that it's a
case of correct numbers producing incorrect recommendations.
Fair summary. His remark about the use of english is a good
example. Is it a problem? Yes? Can anyone see a solution? No (and he
does not suggest one).
More on this after reading some more.
Take care,
John Day