12 May
2008
12 May
'08
5:30 p.m.
--- Brad Parker <brad at heeltoe.com> wrote:
Ray Arachelian wrote:
AppleTalk (in its various forms (LocalTalk, EtherTalk, and even TokenTalk)
worked quite well for its INTENDED purpose. When Apple went from "Phase 1" to
"Phase 2", it was because they realized at the time that routing packets were
really clogging the system. The internal AppleTalk network Apple was using at
the time would be passing around routing information on an idle network for
over 10% of the traffic. It was simply getting out of hand. There was all
this traffic at odd hours in the late night/early morning. "Phase 2" helped
quite a bit as it narrowed down the traffic necessary to get the job done.
Apple's internal network had LOTS of zones in it (all over the world), and even
more nodes (they have lots of desks!). Before the earthquake every was always
running around trying to find out who unplugged what (usually several cubes
away under some desk). It was really bad. I was working on A/UX (Apple's Unix
at the time, pre Linux!) networking in the AppleTalk area, and the idea there
was to get printing to function (it did). Most of AppleTalk was a big push to
get "plug & play" to work, and lots of effort was expended to this goal.
It
did work quite well, but if one was to administer a network, the "plug &
play"
aspect was almost counter to good administrative practices. The largest
problem was naming. Plug & Play implied local naming, and good administrative
practices indicate a central naming scheme.
Oh, well. For the most part, Apple's networking was MUCH easier to use than
the others at the time, if they existed at all. I've even got a couple of PC's
connected to an AppleTalk server (there is code in Linux for it). If you want
to run an understandable networking under MS-DOS, AppleTalk for PC's isn't that
bad. A far sight easier than Novell's stuff (memory footprints aside).
I note that even today, network printers have AppleTalk in them, and will still
connect to ancient 68K based Macs (like my Quadra 840AV). No reason to change
what is quite functional!
Even today, networking seems to elicit "black magic" practices, but we are
getting better!
--
Sorry,
No signature at the moment.
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now.
http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
It's funny how they use CSMA/CA on this network instead of CSMA/CD, and
when ethernet came out, they continued to use the same protocol until
nearly modern days where they switched to encapsulating Appletalk
packets inside TCP/IP frames.
In fairness it would be proper to separate the mac (physical) layer from
the network layer.
All the network layer needed was unreliable datagram delivery and
broadcasts. Both Localtalk and Ethernet provide that.
Collision detect was not really possible on localtalk, unlike ethernet.
But I would argue that CA and CD are refinements which affect maximum
utilization, not gross access to the wire. And both rely on an
appropriate back-off algorithm to work correctly.
(it's fun to go read the original AlohaNet papers and the DIX blue book
for more on that subject)
I was almost like they were re-inventing Ethernet.
:-)
When Sidhu started on that project ethernet was still pretty nascent.
And ethernet was thought of as too expensive and complex (which was
true, I think, when concidering a classic macintosh).
And at that time there were many many protocols-of-the-day. It wasn't
clear at all which would be dominant. IPX, DecNet, TCP/IP, SNA, ISO ...
I suppose they
could have actually ran ethernet over phone cables, but probably someone
thought they had a better way to control collisions; perhaps they did.
I think it was a cost issue. You can't beat the SCC's fm mode (and the
internal pll). They added a little RTS 'beacon' to reserve the wire
which was part of their 'collision avoidance' scheme. It worked pretty
well if you dedicated a cpu to it.
Of course there was nothing but trouble with early
Ethernet switches and
EtherTalk since they continued to use CSMA/CA for quite some time.
I think that was a different problem; Appletalk V1 is very "chatty"
and uses broadcasts a lot. Many early networks were bridged and
this caused problems. I could go on and on, but localized broadcast
based discovery protocols don't work well over long haul bridged
networks, nor do broadcast based distance vector routing protocols
(i.e. NBP and RTMP).
heh. I once crashed (I kid you not) several hundred VAX's with a single
NBP broadcast packet. Bridged networks are not my friend.
I guess there's nothing quite like the not
invented here syndrome.
Certainly some of that, but more in Appletalk v2 and than original
Appletalk. To my mind the original appletalk was simple and elegant - a
good fit for the system...
-brad