dd@ariel.unm.edu (04/12/89)
At UNM, our network is organized as a backbone network (broadband, with each active 6 MHz channel functioning as a 5 Mbps CSMA/CD "Ethernet like" data channel), serving as in inter-building transport for Ethernets. The Ethernets are "glued" to a particular broadband channel in a MAC filtered fashion - no flames please, TCP/IP isn't everywhere it should be yet, and we have a very cooperative user community. As we move into distributed image processing (expected to occur in the next 2~3 years), I need a better backbone. Also, the cost of extending our private CATV system is now (finally!) higher than the cost of extending an optical fiber backbone. And there are other advantages (which you all know more about than I do, in all likelyhood). So fiber it is. And of course, the "right" way to go is to install an FDDI compatible wiring plant, and drive it with FDDI-non-compliant electronics, temporarily. FDDI is logically a token ring system, but the wiring plant may optionally be "star-shaped", i.e., go out to a building, and come back to a "wiring center", and go out to the next building, and come back in,... and so forth, ad nauseum. My question, addressed to anyone who has a solid argument, but particularily to folks who might have installed an FDDI compatible wiring plant is: What is the best way to go? Should I build a ring, or a star-shaped ring, or a ring of rings, or a ring of stars, or what? I realize that the ultimate decision will depend on the particulars of our campus, but if anyone can think of factors that I ought to consider, please let me know. Thank you all for your help! Don Doerner dd@ariel.unm.edu University of New Mexico CIRT 2701 Campus Blvd, NE Albuquerque, NM, 87131 (505) 277-8036
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (04/13/89)
In article <4824@charon.unm.edu>, dd@ariel.unm.edu writes: > > My question, addressed to anyone who has a solid argument, but > particularily to folks who might have installed an FDDI compatible > wiring plant is: What is the best way to go? Should I build a ring, > or a star-shaped ring, or a ring of rings, or a ring of stars, or > what? I realize that the ultimate decision will depend on the > particulars of our campus, but if anyone can think of factors that I > ought to consider, please let me know. > > Thank you all for your help! > > Don Doerner dd@ariel.unm.edu [This turned out rather long.] We at Boston University installed Pronet-80 in '87 on a fiber optic cable plant and we are writing the contract for the contractor to install a new fiber plant in new conduit to extend our fiber network for Pronet-80 and other uses. Perhaps what we did in our pilot installation and what we designed for the extension would be instructive. As you know, Pronet-80 is essentially the same as FDDI with the important exception that FDDI has Station Management. Pronet-80 has no management of any kind. The cable plants are the same. Other uses for fiber include f/o Ethernet and sometime in the future, telephone services might be useful on campus over fiber. I am including single mode fiber in the entire cable plant so I can distribute external services anywhere on campus as needed. When we designed our pilot cable plant, we installed the fiber in a big ring around a street in our campus. We installed patch panels in the basement of each building and we ran 24 fibers between each patch panel. We could install a f/o interface in each building and the distance between f/o interfaces would be a short hop on the trunk between panels. Of course, for those buildings with no f/o service, we patched through the panel to get to the next panel. I think we have about a dozen patch panels, but only seven routers. That means that there are a lot of patch connections in some of our links and most of our loss budget is in the connectors. This is no problem over a short distance. But then in the process of debugging a problem with the receivers in the f/o interfaces, we ran up against the problem of too much loss in some links and we had to go back and take out some patch cords to reduce the loss. Still not really a problem. Then we had a need to do some f/o Ethernet extension to solve a problem we had to grandfather, so we needed some fiber with a different topology. Still workable, but getting a little kludgy. Now, scale up the distance, number of buildings, etc and you can see that this topology design begins to get unworkable very fast. Our extension project triples the geographic scope of our fiber cable plant with almost no expansion of our router base. That part of campus is less compute-intensive today, so the fiber is really more for the future than to solve today's problems. However, our admin computing people will be joining this extension, so we will have our hands full. Anyway we have a big sparse network and the need for some low budget networking. Not every Mac cluster can afford a FastPath and a $25k p4200 with Pronet-80. :-) After much looking around at other designs and much thought, etc I came up with the multihub star topology as our preferred solution. A full blown star was impractical because we didn't have a logical place to put the hub and a ring topology just wasn't practical in terms of providing low loss paths where we wanted them in our sparse matrix. We defined three hub sites in our extension: one on each end for service coverage and further extension, and one roughly in the middle. We joined one end hub to our pilot plant thru our computer center. The hubs "tile the plane". That is, they each have a service area that covers the whole extension area without overlap. Each hub serves from three to four buildings. The building service is star configured from each hub to each building in its service area. We are running twelve multimode fibers to each building. The center hub is linked to each of the end hubs with a dual trunk cable, each cable contains 24 multimode and 4 single mode fibers. Unfortunately, I do not have widely separated paths to route each of my trunk links thru, but I recommend physically dispersing your trunk runs as much as possible. I am having them routed through different ducts in the bank and down each side of the vaults, etc to maximize physical redundancy and fault tolerance. You could tie your hubs together in a ring or a matrix, depending on geography and number of hubs. I will end up placing a router in each hub and they will be linked on the unspliced, continuous trunk cables. Then I can install other routers or f/o ethernet extensions from the hubs to the buildings in the service area. I can place most anything in the hub and link it up with fiber to any or every building in the service area. Maybe even PBXs. We designed a broadband extension with exactly the same topology and a trunk amp in each hub. All trunk service is in one of three hub closets. No equipment in any manhole. Now when we plug our f/o interfaces together we just have to be sure that there is enough loss in each path so we don't have to buy attenuators. Our longest trunk run is 2800 feet, I think, so, again, loss is determined by connectors and not fiber loss. I am sure we will like our multihub topology much better. I hope the trunk fibers last a while before we have to pull more. BTW, the broadband trunk pulls are going to be very tough, requiring winches and all kinds of special work. The fiber following the same path can be pulled in by one man by hand. You could probably do it yourself. Our broadband trunk is one-inch cable. My tech is very demanding about signal quality. :-) The fiber is about half an inch, feather light and easily flexed. --Kent England, Boston University
morgan@Jessica.stanford.edu (RL "Bob" Morgan) (04/13/89)
As my last act at San Francisco State University in 1987, I designed a fiber plant to support FO Ethernet (using Codenoll xcvrs) now and FDDI very eventually. I ended up with something that is a lot like what Kent just described for his new net at BU: a set of interconnected stars. In this case it was three stars, each with a "passive" center, each serving four to six buildings with 12 multi-mode fibers per building. The stars were connected together with 12 more fibers (probably should have been more). This allowed a logical design that is currently very Computer-Center-centric (it's at one of the stars), using Ethernet bridges to link the stars, while allowing ring or whatever eventually. (Stanford, BTW, is a living museum of cable and signalling methods, and probably not a model for anyone's planning, at least for baseband applications.) - RL "Bob" Morgan Networking Systems Stanford
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (04/13/89)
In article <1507@Portia.Stanford.EDU> morgan@Jessica.stanford.edu (RL "Bob" Morgan) writes: > >(Stanford, BTW, is a living museum of cable and signalling methods, >and probably not a model for anyone's planning, at least for baseband >applications.) > I have this theory of installed wire as a complex system. Once your installed base of wire exceeds a certain critical threshold (no one ever pulls anything out of his wire "museum"), the cabling takes on new and complex behaviours. The wire becomes alive, capable even of eating technicians who foolishly venture in the zone above the ceiling tiles. We lost two techs. We left the ladders where they were and the ceiling tile off, hoping the wire beast would at least spit out the bones, but it never did.
dsmith@oregon.uoregon.edu (Dale Smith) (04/13/89)
In article <1507@Portia.Stanford.EDU>, morgan@Jessica.stanford.edu (RL "Bob" Morgan) writes: > As my last act at San Francisco State University in 1987, I designed a > fiber plant to support FO Ethernet (using Codenoll xcvrs) now and FDDI > very eventually. I ended up with something that is a lot like what > Kent just described for his new net at BU: a set of interconnected > stars. In this case it was three stars, each with a "passive" center, > each serving four to six buildings with 12 multi-mode fibers per > building. The stars were connected together with 12 more fibers > (probably should have been more). This allowed a logical design that > is currently very Computer-Center-centric (it's at one of the stars), > using Ethernet bridges to link the stars, while allowing ring or > whatever eventually. At the University of Oregon, we have taken a slightly different approach and installed a "star of rings" rather than the "ring of stars" described by Kent of BU and Bob of Stanford (network of SFSU). We have three interconnected rings that cover most of our 1 square mile campus. From our Computing Center, which is a node in one of the rings, we run point-to-point to the other rings (thus the star of rings). As Kent points out, you must be careful about loss in buildings with no active devices. You should engineer for losing 1dB per building that you patch through. Actual practice should be somewhere closer to .5dB, but you should engineer for 1dB. With FDDI, this can become a big problem fast. FDDI has an 11dB loss budget. If you want to survive a failed station that has gone into bypass, then you need to figure you'll lose 4dB through the bypassed station, giving you 7dB. But, you have to spit the 7dB between runs on each side of the failed station, giving you a 3.5dB budget between any two stations. Figuring 1dB per inactive building plus a little loss in the cable, you can see you don't want to build huge rings. Note that I have been very conservative in the figures above and you can get by with a lot more. Special point-to-point applications also pose problems in a ring topology. You can potentially end up using lots of fiber for simple links. We have carefully planned where our star presences are so that we can minimize the amount of fiber required for special applications (we hope). A drawback of running FDDI over a physical star is that a single media failure can partition your network, whereas no single failure can partition your FDDI network if you have a true ring. Dale Smith Internet: dsmith@oregon.uoregon.edu University of Oregon BITNET: dsmith@oregon.bitnet Computing Center UUCP: ...hp-pcd!uoregon!dsmith Eugene, OR 97403-1212 Voice: (503)686-4394
bud@ut-emx.UUCP (04/13/89)
In article <29516@bu-cs.BU.EDU> kwe@buit13.bu.edu (Kent England) writes: >In article <1507@Portia.Stanford.EDU> >morgan@Jessica.stanford.edu (RL "Bob" Morgan) writes: >> >>(Stanford, BTW, is a living museum of cable and signalling methods, >>and probably not a model for anyone's planning, at least for baseband >>applications.) >> > I have this theory of installed wire as a complex system. >Once your installed base of wire exceeds a certain critical threshold >(no one ever pulls anything out of his wire "museum"), As a former curator of the Stanford wire museum I can second the observation that no one ever retires a network system that runs, no matter how poorly designed or implemented it is. There ain't no such beast as a "temporary" network. Network designers be warned, if you install a quick and dirty hack (like we did at Stanford), odds are it will become a permanent embarassment.
morgan@Jessica.stanford.edu (RL "Bob" Morgan) (04/14/89)
Kent writes: > The wire becomes alive, capable even of eating technicians who > foolishly venture in the zone above the ceiling tiles. We lost two > techs. We left the ladders where they were and the ceiling tile off, > hoping the wire beast would at least spit out the bones, but it never > did. Ah, there's your mistake: true cable techs have no bones, the better to sneak cables thru skinny conduits (though, of course, *real* cable techs don't use conduits, or plenum-rated cable). Your report of disappearing technicians adds fuel to my conjectures about the true composition of Yellow-77 . . . *8^)* - RL "Bob"
ciriello@lafcol.UUCP (Patrick Ciriello II) (04/14/89)
We have just finished designing a cabling plant that divides that campus buildings into 16 logical rings, all connected by fiber, and a backbone that connects all the rings. Each building will get 18 fibers. This will allow for a main path (with built in backup path), a full backup set of fibers (used 4 so far), 2 for another net to run concurrently (if neccessary), 2 for its backup, and the rest to be used for any future needs (including telephone, video, security, direct links between certain dept.'s, etc). We are linking 44 buildings, and the total length of fiber is about 5.85 miles.