Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uunet!aplcen!uakari.primate.wisc.edu!brutus.cs.uiuc.edu!wuarchive!psuvax1!psuvm!USU!JRD From: JRD@USU.BITNET (Joe Doupnik) Newsgroups: bit.listserv.novell Subject: Re: ETHERNET PROBLEM Message-ID: Date: 1 Feb 90 14:50:00 GMT Sender: Novell LAN Interest Group Reply-To: Novell LAN Interest Group Lines: 51 Approved: NETNEWS@PSUVM Gateway X-VMS-To: IN%"NOVELL@SUVM.BITNET" Dave, Ethernet cabling discussion. Normally the separation between thin or thick wire cable taps is 2.5 meters. The reason is to break up the small reflections from each tap and is related to the propagation velocity of the signal in the cable. The final terminator can be attached to the T connector on the final PC, no extra short wire is needed. If you use one of the vendor's "bright ideas" on longer length runs then beware that all (and I do mean all) the boards must be set for the same peculiar signaling method because otherwise the collision detection circuitry will declare collisions nearly all the time (the signal level is outside the standard ranges for 0V and -2V levels). Try this: regular signaling, put a scope on the last PC, then remove the terminator. The signal level jumps up by a factor of 2 from the reflection. That's the level for these extended length ideas. The official max thin wire segment is 185 meters and is based on the attenuation of the signal through typical cable. With better quality foam dielectric and double shielded thin wire, such as made by Belden for Digital at $210/1000 feet, that attenuation is slightly less and hence the cable can be lengthened by say 10% or so (which we all do anyway, given how wires get run). What we must never do is make the electrical length of the cable longer than 256 bit times (Ethernet 512 bit slot time / 2) because efficient collison detection requires that the sender be able to hear a collision while it is transmitting. The worst case senario is PCs at the far ends transmit almost simultaneously, A going before B by the cable propagation delay. B starts nanoseconds before A's signal gets to it, B sees a collision and quits, A must see that tiny collision fragment and also declare an error BEFORE it finishs the current packet. With the 512 bit min packet size that sets the max cable length to 256 bits. In other words, yes, a little stretch is ok but more is worse. RS232 and Ethernet have no, repeat no, relation about lengths. The underlying princples are different. Ethernet repeaters ($$$) help with long lines, and help in a subtle way of cleaning up gitter of the individual bits from reflections. Two of these fellows per cable, max. After that one needs a MAC level bridge ($1800, or an el cheapo PC with PCBridge software and consequent management headaches). Finally, just a word on cable types for those listening to previous discussions on that matter. RG58 is a generic identification, like Chevy. Real plain RG58 is 53.5 ohms impedance stuff; not what we need. RG58A is 50 ohms. Cable of the RG description is ok for short runs but specially made low loss cable for Ethernet is better all around. As mentioned above, DEC sells the grey jacket, double shielded, stranded center conductor, foam dielectric thin wire, made by Belden, for about $210/1000 feet. Good price for a good cable. Another point is that the connectors should be the crimp type, never the "easy on" or "twist on" or other trash, and the center conductor pin gets crimped too. A good crimp tool for RG58 gauge cable costs around $40 to $80; expensive but worth it, especially the latter kind. Wire is the cheapest part of the whole installation so don't cut corners there. Connectors are expensive ($2+ apiece) but the good ones don't require servicing at embarassing times. Joe D.