On Wed, 2001-11-07 at 06:23, Eric L. Strobel wrote:
>
> Thanks for the explanation. Question, though... It seems to me that, given
> the bandwidth of even 10BT, every five seconds is an eternity. I guess I've
> never understood why this was a problem. We're talking, what, a few
> THOUSANDTHS of a percent of the bandwidth per machine pair?
Ah, but no.
Each machine that offers Appletalk services does an Ethernet broadcast
of nontrivial size every five seconds. That is, it sends a message that
lights up the wire to -every- other machine. Mainframes, Pilot cradles,
anything that is physically on the same Ethernet segment, it all sees
the broadcast. Anything that was talking at that point has to back off
and retry because of collisions.
Now you put two machines on there. Assuming best-case scenario, now
this happens from one machine or the other every 2.5 seconds. Worst
case? They happen to want to do that at the same time. And collide
with -each other-, thereby prolonging the spammy event.
Now put fifty machines on the same segment, and start to feel the pain.
> Maybe I've
> never worked anyplace big enough to have a problem (I realize the traffic is
> going to go like N^2 IF every Mac has file-sharing on, otherwise isn't it
> just # AppleTalk printers X # Macs?).
See above. It's not a point-to-point connection between AppleTalk
devices. If we could do that, we wouldn't need to be advertising that
we exist every few seconds. AppleTalk has no stateful concept of who's
on the wire, no name servers, no browse masters, nobody collecting the
state of the network. Each machine keeps its own state, based on
hearing these broadcasts. So we have to keep advertising to the entire
network, in case somebody just came alive that we don't know about.
It's really really brute force and really really ugly, but it does make
for the ease-of-use Apple promised. Oh, and there's a whole chapter in
this story about how AppleTalk originated to live on a big
interconnected mesh of serial lines, so grafting it to a smarter medium
like Ethernet was basically an exercise in horrible compromises.
> How does this compare with the
> problem that print spoolers cause? (You know, sending those 20 MB
> Powerpoint files over the network TWICE, once to the spooler, then from
> spooler to printer.)
On a switched Ethernet network, that 20MB file only lights up the wires
between participating parties. No other machine even knows it's
happening; nobody else's bandwidth is taken up. Unless an Ethernet
broadcast happens, like a machine coming up and needing a DHCP address.
Or a Mac realizing that its friends might be lonely since it hasn't
spoken to them in five seconds. Then there's collisions and we have to
retry frames.
> {I suppose this is wandering OT... I'll shut up now.}
I think discussions of AppleTalk, the Newton's original networking
scheme, are less OT than discussions of Star Trek....
-- R Pickett The people that once bestowed commands, consulships, Hayseed Networks legions, and all else, now meddles no more and longs emerson_at_hayseed.net eagerly for just two things -- bread and circuses.-- This is the Newtontalk mailinglist - http://www.newtontalk.net To unsubscribe or manage: visit the above link or mailto:newtontalk-request_at_newtontalk.net?Subject=unsubscribe
This archive was generated by hypermail 2.1.2 : Sat Dec 01 2001 - 20:02:18 EST