ARCHERB@VAX1.UMKC.EDU (Barry Archer) (01/24/91)
We're looking into finally expanding our network throughout the campus. UMKC is an urban campus with a number of buildings to cover and ancient, meandering conduit. Once we get off the backbone, we've been looking at using a combination of ethernet concentrators (such as Chipcom's, which allows an ethernet star config) and cisco IGS routers. So the route to a building might look like: Backbone -- Big_Router -- ethernet_concentrator ! building A | IGS -- ethernet_concentrator ! building B | IGS -- ether_concentrator ! building C The links between concentrators & routers would be ethernet over fiber, with the max distance between the two less than 1km. The concnetrators would give us a wiring hub within each building. Does this appear reasonable? How deep can I go with the routers? I'd like to keep any one host from having more than 4 hops before getting off campus, since it takes another 4 hops to get to the NFSnet. Is there a good rule of thumb? Any comments welcomed ( I think? :-) ). Barry Archer, UMKC NOC archerb@gawain.umkc.edu archerb@vax1.umkc.edu archerb@umkcvax1.bitnet
byczynsk@udel.edu (Larry Byczynski) (01/24/91)
In response to your question, does it appear reasonable, if the big router is yours, why do you need the IGS's? Also, keeping within the 802.3 specifications of 1km for fiber links, is not necessary when extending an ethernet from one host to another. It is a matter of propegation delay. For more info, send mail to byczynsk@huey.udel.edu.
kwe@bu-it.bu.edu (Kent England) (01/24/91)
In article <31735@boulder.Colorado.EDU>, ARCHERB@VAX1.UMKC.EDU (Barry Archer) writes: > > Backbone -- Big_Router -- ethernet_concentrator ! building A > | > IGS -- ethernet_concentrator ! building B > | > IGS -- ether_concentrator ! building C > I would recommend avoiding the extra hops through the IGSs, the extra expense, and the extra configuration management. There are many advantages to minimizing the diameter of the Internet and what you do on your campus is very important to keeping that diameter small. If you plan on using IGRP or RIP, the smaller the diameter the faster the routing convergence which I happen to think is important. If you have sufficient capacity in Big_Router or Big_Routers I would suggest you consider extending an Ethernet interface out to each of the concentrators with pairs of fiber optic transceivers. You must be cautious about violating the delay spec for whichever Ethernet extension you use, because delay violations will result in undetected collisions which are correctable, after considerable delay, only by higher layer protocols. If you wish Ethernet extensions of arbitrary delay, you can talk to me or Kirk Lougheed at cisco about the two-interface hack we did for NEARnet microwave Ethernet links, but it requires a router on each end. I'm not a fan of putting all my eggs in one basket; at BU we use a small set of routers on a high speed backbone with Ethernet f/o extensions to buildings throughout largish service areas. This seems like a reasonable compromise among cost, reliability, ease and performance. If you think FDDI is too pricey or dicey as your backbone, you could use a couple or three Ethernets as backbone among the set of Big_Routers servicing your buildings; ethernet interfaces are cheap on cisco routers; use them liberally. --Kent