Re2: VERY Large AppleTalk netw

["Oppenheimer, Alan" <[e-mail]>]

Tom Evans writes,
 
> IMHO (which may be wildly wrong and/or outdated) AURP requires
> "full-mesh" connectivity via "tunnels".
 
No comment on the wildness or outdateness of your opinions, but they are wrong.
AURP does not require full-mesh tunnels.  Partial connectivity is allowed and
is in fact viewed by most as a feature.  On the AppleTalk Internet today, we
have 15 schools or so, but no where near 14*14 AURP-Tr connections.
 
> Each "tunnel" needs to be independently maintained by "tickle"
> packets from both ends every 30(?) seconds. AURP uses a "reliable"
> transport system requiring acknowleges (2 packets per tickle).
 
AURP-Tr (the default transport mechanism for AURP) does use a reliable tickle
mechanism, which does require two packets per tickle.  IN NO WAY, HOWEVER, ARE
TICKLES REQUIRED, especially not every 30 seconds.  The tickle time is totally
configurable, and can be set to infinite (no tickles required ever on a "quiet"
tunnel).  AURP does require a tickle be sent before an AppleTalk data packet if
the other side hasn't been heard from for the Tickle-Before-Data time (2
minutes).  The Apple Internet Router, Apple's implementation of AURP, has its
default tickle time set to once every 90 seconds.
 
> 1000 AURP routers means 1000 * 999 / 2 tunnels (499,500).
 
AURP's design center was on the order of 1000 internets or so, and clearly this
many tickles would have been totally unacceptable.  On a tunnel of this size,
you would probably configure all the routers' tickle intervals to infinite, and
thus tickle traffic would be zero except on those connections which are active,
and on those connections it would be a small fraction of the data being sent.
 
> This leads to the following routing traffic for a 1000-router internet:
 
Actually, the real routing traffic is not tickles, but the propagation of
routing information.  If the internet is totally stable, there will be no
routing information to be propagated, but usually there will be an update once
in a while.  If you really wanted to have a fully-connected tunnel of 1000
internets, sending these reliable updates would probably use as much or more
bandwidth as tickles.  Again, however, AURP was designed to be scalable, and by
tuning the update interval appropriately, such a tunnel could be built
(admittedly with less dynamicism than a standard AppleTalk internet).  Apple
provided the following lines in a chart to MacWeek, who then mangled the whole
chart before publication.
 
        Protocol        Bytes/sec       % of T1 line
        ---------------------------------------------
        RTMP            2,300,000       >100%
        AURP            463             .2%
 
The rest of the chart is available if anyone is interested.  It assumes one
update every 24 hours and use of the tickle-before-data feature.
 
> I guess then that AURP is not (was never?) meant for
> in-campus-connectivity, but only for a small number of
> between-campus connections.
 
It sure is/was!  See above.  AURP was designed, in consultation with our
developers, universities, and customers to support tunnels into the thousands
of internets.  We did draw the line somewhere under 10,000, which some day may
come back to bite us, but I'm actually looking forward to having to deal with
this problem!
 
I also want to take this opportunity to apologize for not having the
Information RFC on AURP available yet.  Hopefully the first draft will be out
next week, in time for Mactivity.