mckenzie@LABS-N.BBN.COM (Alex McKenzie) (03/31/88)
I believe the product in question was designed with the following goals: 1) Use Ethernet interface, so customer could use any standard Ethernet card in the attached equipment. 2) Use FO transmission medium for less bulk and greater resistance to environmental electrical issues (noise, grounding, wiretapping) The ring topology was chosen as the easiest way to get the signal around to all the stations with FO. The above is based on my fuzzy recollection of a conversation with the vendor a few years ago. As I recall, the taps are active, which makes them cost more. Collisions happen within the taps, where the signals are regenerated, rather than on the "wire". This means that the Ethernet cable length restrictions don't apply, since those restrictions are required only to insure that if one station sees a collision, all stations will. I think it also means Ethernet packet size restrictions are not required, but since the intent was to use standard Ethernet boards, this restriction is retained. If you want to use Ethernet and FO, another alternative is the FO "star couplers" sold by Siecor and Codenoll, among others. In this approach the light signals from a number of fiber runs are combined optically in a fused glass central location. We are using one of these at BBN as the "backbone" to which each of our many coax Ethernets is connected (via a bridge). The biggest disadvantage to this approach is that any changes are harder than one would like, due to the careful consideration that has to be given to the light level on each arm of the star (the signals have to be well enough balanced so that a collision between the strongest and the weakest is noticed, and the signal input on each arm has to be strong enough so that 1/n-th of it is strong enough to be detected by the receiver which is fartherst away). A third approach is to install coax Ethernet in every machine cluster and bridge each one to one of the "Ethernet in a box" devices; the bridges can be connected via FO cable. This uses FO for the long runs, while keeping the cost of attachment of individual devices low. This approach is currently my personal favorite. Hope this helps, Alex McKenzie
LYNCH@A.ISI.EDU (Dan Lynch) (04/01/88)
Alex, Thanks for the excellent tutorial on the difference between layer 2 and layer 1! (or have I messed up the numbers???) It continues to amaze me how innovative people can figure out how to maximize things for new situations. In the last few weeks I have seen references to "brouters" (bridge/router -- level 2/level 3 animals). I get the feeling that the primorial ooze of networking (or distributed computation) is still in the fermentation stage and is not "ready for drinking". Anyway, Steve, you got "lucky" ("smart" is more accurate because Steve know there was a resource out there, somewhere) because Alex chimed in. Dan -------