glen@aecom.yu.edu (Glen M. Marianko) (10/25/89)
There is a stated limit of (approximately?) 30 nodes on an localtalk segment. I assume when they say nodes, they really mean localtalk adapter boxes that plug into the localtalk cable with a short drop cable to the actual node. Is this limitation the same in the phone-net type topology? What is the reason for the limit? Reflections from the boxes? I assume active devices like Tops Repeater can extend this until you reach the 254 node limit and then you use something like an Interbridge or Shiva Netbridge? -- -- Glen M. Marianko Manager, LAN Services Glasgal Communications, Inc. 151 Veterans Drive Northvale, New Jersey 07647 201-768-8082 glen@aecom.yu.edu - {uunet}!aecom!glen (Courtesy of AECOM & unaffiliated)
tom@wcc.oz (Tom Evans) (11/10/89)
In article <1989Nov2.013604.24598@polyslo.CalPoly.EDU>, dorourke@polyslo.CalPoly.EDU (David M. O'Rourke) writes: > tom@wcc.oz (Tom Evans) writes: > >There are better reasons to use LocalTalk routers before that limit is > >reached. <rubbish about dividing the bandwidth removed> > > I don't understand why you divide the bandwidth. Each node doesn't get > assigned a chunk of the bandwidth. When one node is using the network it > has all of the network bandwidth availible to it for the duration of that > particular packet's transmission time. The problem with lots of nodes > on any contention {sp??} based network is that as you add nodes to the > network the probability of a collision, and hence a re-transmission, goes > way up. When you get collisions you have to backoff and retransmit which > waste more bandwidth and lowers overall throughput. Two points. With a high speed network (Ethernet), it takes a fairly busy host to grab more than a fraction of the bandwidth, and thus a lot of hosts till it's "full". I can grab 54% of a LocalTalk network with ONE MAC SE! I just copied a 1.9Mb stack from a server and it took 127 seconds - that's 14.7kbytes/sec. LocalTalk's theoretical max without ANY OVERHEAD is 28kbytes/sec. If you have two Macs doing file transfers, Application launches or heavy printing, they start stomping each other. You're sharing bandwidth REAL EARLY. The only soltion to this is to add LocalTalk Routers where needed. With Ethernet you do get a loss of effective bandwidth as the traffic and the resulting collision rate go up. With LocalTalk, there are no "detected" collisions. Collisions only occur between the short "RTS" packets which are 3-byte (ok, 8.5 bytes on-the-wire) packets. A data packet is up to 600 bytes, so you don't lose as much to collisions as you'd expect. Yes, I agree totally about the number of connectors in a 30-node network. --------- Tom Evans tom@wcc.oz.au | Webster Computer Corp P/L | "The concept of my 1270 Ferntree Gully Rd | existence is an Scoresby VIC 3179 Australia | approximation" Australia | 61-3-764-1100 FAX ...764-1179 | D. Conway
tom@wcc.oz (Tom Evans) (11/10/89)
In article <7054@quick.COM>, srg@quick.COM (Spencer Garrett) writes: > There are 4 reasons I can think of for limiting the number of nodes > on networks in general, and one additional reason for limiting them > on a Localtalk net. > > 2) Signal attenuation - Each transmitter on a network has to drive > *every* receiver on the network. Eventually the signal > levels get so low that they are indistinguishable from the noise. Only if you're driving a star configuration. The receiver should have a high-resistance and a low-capacitance. It doesn't "eat" the signal. > 3) Contention resolution - ... > Binary exponential > backoff is the most efficient, but if you don't know where to > truncate the series then the waits can approach infinity LocalTalk doesn't use it. It generates a "random" backof with an upper range determined by the last 8 attempts. It ranges from 0 to 1,500 uS. Max wait is 400 + 1500, which sounds a lot, but is only 55 byte-times. > 5) Addressing - Localtalk uses 8 bit addresses, and probes for an > available address each time a node is made active. As > the fraction of available addresses which are actually > in use rises this process becomes more and more painful. It probes for the last one it used. It degenerates to statically-addressed. You'd have to have more than 100 nodes to have this problem, and by the time this becomes painful, you'd be dead from all the other limits. --------- Tom Evans tom@wcc.oz.au | Webster Computer Corp P/L | "The concept of my 1270 Ferntree Gully Rd | existence is an Scoresby VIC 3179 Australia | approximation" Australia | 61-3-764-1100 FAX ...764-1179 | D. Conway