morrison@thucydides.cs.uiuc.edu (06/14/90)
Hello, Adding 'extentions' to the T connector is something that many people ask about. It is a bad idea, unfortuately, people try it and it works, so they leave it. Ethernet is a pretty robust spec, so you can do all sorts of 'bad' things (like exceeding length limitations and adding T's etc), and it can take it (up to a point), but you should realize that you are pushing your luck, and if you keep doing it sooner or later your luck will give out. Why are T extentions bad? An ethernet should be thought of as a 'light pipe'. That is the electrical signals travel up and down that cable at the speed of light. Just like light, when it hits the end of the cable it is reflected back, and when it hits a T connector, some is reflected back, some goes down each branch of the T connector. Now normally we don't have to worry about the ends because the termination resistor is chosen so that all the signal is absorbed. Thus we only have to worry about reflections at the T connectors. (1.0) --> ---------------+---------------- | | Now when a signal hits a T connector, 2/3 of the signal goes down each branch, and -1/3 gets reflected back to the source. Now one of these branches is the rest of the ethernet, and one is connected to the computer. The Voltage at any point is simply the sum of all the signals that have passed that way. Notice since the voltage at the T can have only one value, it had better be the same regardless of which branch we compute it on. If we compute it on the incomming branch it is 1 - 1/3 = 2/3 which checks since that is the voltage along each of the outgoing branches (1.0) --> <-- (-1/3) (2/3) -> ---------------+---------------- | 2/3 | | | v Now let us assume there is a host on one end of the T. Now this host looks like an OPEN circuit. Open circuts reflect back all the signal so a small time latter the signal looks like (1.0) --> <-- (-1/3) (2/3) -> ---------------+---------------- | 2/3 ^ | | | | v 2/3 Now this reflected signal hits the T and is reflected, 2/3, 2/3, -1/3 so it looks like this (1.0) --> <-- (-1/3) (2/3) -> <-- (4/9) (4/9) -> ---------------+---------------- | 2/3 ^ -2/9 | | | | | v 2/3 v And after another reflection and another trip through the T it looks like. (1.0) --> <-- (-1/3) (2/3) -> <-- (4/9) (4/9) -> <-- (-4/27) (-4/27) -> ---------------+---------------- | 2/3 ^ -2/9 ^ 2/27 | | | | | | | v 2/3 v -2/9 v Thus the voltage at the host after a 'long time' will be 2/3 + 2/3 + -2/9 + -2/9 ... = 4/3(1 - 1/3 + 1/9 ...) = 1 and the signal continuing down the line after a 'long time' will be 2/3 + 4/9 - 4/27 ... = 2/3 + 4/9(1 - 1/3 + 1/9 ...) = 1 Which is exactly what we would expect, after a 'long time' the open circuit has no effect on the signal traveling down the cable. Now normally it only takes about 5 or 6 reflections for the voltage to stablilize, and if the distance from the T to the host is small, then it can be thought of as being instantaneous. However if the distance from the T to the host is say 1 Meter, light will take about 6 ns to travel from the T to the host, and thus about 30-40ns to stablilze. Now the bit time for ethernet is 100ns. With a 1 meter extention, it takes about 1/3 that to stablize, so it will probably work, but the signal will be noticably degraded. Also note that this effect is cumulative. If another host down the line has a 1 Meter T, it will take the degraded signal and degrade it further. Thus an ethernet can probably only tolerate at MOST about 5 meters of cumulative T length before the signal is degraded to the point where it becomes intolerable. Note that this can cause real headaches later on because of the distributed nature of networks. One person adds a short T extention and now some hosts can't communicate relyably to others, but since you didn't know about this other person adding the T extention it takes you a week of debugging to figure it out. Thus the moral of the story is 'don't do it' and if you MUST, make them as SHORT as possible, and keep a clear log of all such spec violations, so that if you have problems, you know where to look first. Vance