BALAMUT%OLYMPS@engvax.scg.hac.COM ("Morris Balamut - The Dungeon Master 513-5829", 213) (03/21/88)
I recently received the following answer to my request for help/information to solve a problem of a growing number of Macs on ethernet. >From: morganr@SDSC.BITNET >Subject: Getting Loads Of Mac's On Ethernet > >I thought that using Appletalk zones was the solution. Have you checked >with Kinetics 415-947-0998 or Pacer Software 617-898-3300 (Bill Gill)? Apparently I did not make the problem clear and I appologize. We are faced with a growing number of Macs that want to connect DIRECTLY to the ethernet. You CAN NOT run multiple zones over the ethernet. This results in a limit of 254 devices that can be connected to the ethernet and run appletalk. This would be fine if the only devices were Kinetics Faspaths or equivalent. Unfortunately many of the users want to be able to run DECnet and Appletalk. This REQUIRES that the Mac be directly connected to the ethernet. We are currently investigating numerous solutions both hardware and software. Until we have a final solution we have adapted a policy of restricting the Macs from connecting to the backbone ethernet. This is NOT satisfacory (for anyone). If anyone has any experience with this and can make some suggestions please do so. I am sure that this problem is not limited to our organization. I will be happy to collect the responses and produce a summary for the list. I think this will be information that many people will be interested in. In addition I will keep you advised of our progress. Morris Balamut balamut@olymps.hac.com Network Support Group balamut%olymps.hac.com@oberon.usc.edu Hughes Aircraft mbalamut@ecla.usc.edu --------------------------------------------------------------------------- Disclaimer: The preceeding represents additional random hallucinations of my fertile imagination and may not reflect the opinions of my boss or my boss's boss.
morgan@jessica.stanford.EDU (03/22/88)
Regarding the 254 node limit on an AppleTalk network: We talked to some people at Apple about this a few weeks ago. While no definite statements were made one way or the other, the impression I got was that they had heard of this problem and were concerned about it, but had no immediate solution. A proper fix involves some real violence to the existing AppleTalk scheme. They admitted that AppleTalk has outgrown its original design goals (as have IP and DECNet, to name two others), and some hard thinking has to be done about where it goes from here. I think the implication is that the fix to this problem would only come along with a top-to-bottom redesign. In the mean time, many of us think it's crazy to run a single logical Ethernet with hundreds or thousands of nodes. Segmenting with routers is the way most people appear to deal with these issues in the IP and DECNet worlds; unfortunately, there's no high-performance AppleTalk router yet. Bug your router vendor to support such a thing. We bug ours. - RL "Bob" Morgan Networking Systems Stanford University
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (03/23/88)
In article <Added.8WFTNUy00UkTIHF050@andrew.cmu.edu> morgan@jessica.stanford.EDU writes: > >AppleTalk has outgrown its original design goals (as have IP and >DECNet, to name two others), and some hard thinking has to be done >about where it goes from here. I think the implication is that the >fix to this problem would only come along with a top-to-bottom >redesign. AppleTalk is definitely a small network design. I think it's a good design, but it suffers when you try to enlarge it. Problems include: o totally dynamic configuration o liberal broadcasting for names and addresses o time-outs and retrys are just too fast o no internet These are among the reasons I don't think it wise to build campus-wide AppleTalks, but rather integrate into a campus-wide IP network. No reason to reinvent the wheel. Gateway AT to IP. Kent England Boston University
brad@cayman.com (Brad Parker) (03/28/88)
In article <Added.8WFTNUy00UkTIHF050@andrew.cmu.edu>, morgan@jessica writes: > >Regarding the 254 node limit on an AppleTalk network: >... >A proper fix involves some real >violence to the existing AppleTalk scheme. They admitted that >AppleTalk has outgrown its original design goals (as have IP and >DECNet, to name two others), and some hard thinking has to be done >about where it goes from here. I think the implication is that the >fix to this problem would only come along with a top-to-bottom >redesign. > While I totally agree that it make more sense to break up your ethernets, I not sure I understand this statement. It seems that ethertalk nodes could be made "aware" of multiple logical AppleTalk networks on one single ethernet to short-circuit the standard "packet-bound-for-another- network-send-it-to-the-bridge" code. (sorry - I was just hacking some lisp code for nntp/gnews). I.e. it seems that AARP could be changed slightly (or perhaps not at all) to use the 24 unused bits (16 for a network number come to mind). Then all that is needed is a slight change to the code in MLAP to consult a local table of "logical" networks on the local wire. If the network is not in the table, then the packet get's send (as it would be now) to a bridge. (a discusting hack which might work now would be to hack AARP to include networks and then set up a "psuedo bridge" on the ethernet which would retransmit packets sent to it; the psuedo bridge would know which nets where local and which where not.) If my notions are completely wrong, please let me know. Sometimes I have flashbacks ;-) -brad