jeff@eniac.seas.upenn.edu (Jeffrey M White) (04/11/89)
I'm in the process of setting up a larger than recommended passive star
7-10 branches instead of 3-4) using Phonenet type connectors and wiring, and
need some help with the final installation. (for those who may suggest it,
I know about Farallon's Star Controller, but the price is too high (> $1000),
and this installation is sort of a test installation anyway).
My problem has to do with the terminating resistors. For a long time, I
was confused as to whether the terminating resistors were:
1. To absord reflections - in which case you wanted one at the end of every
line, or
2. To keep a constant impedance across the line - in which case you would
only want several of them.
Farallon was nice enough to send me a copy of their Phonenet installation
book, which describes some various wiring configurations. Before I go on,
I should mention that right now, I am actually using Apple Localtalk
transceivers, but using unshielded wire. At first I had about 4 or 5 branches
connected, some with 2 or 3 devices on each, and there was some qualitative
problems (Tops workstations kept on losing contact with their servers). The
reason, I suspected, was that all the Localtalk terminating resistors were
connected. So what I did was to plu some dummy connectors in the 2nd port,
which opened up the resistor. I believe I kept 2 or 3 TR (term. rest.) in
the circuit). That seemed to improve things until I started to add more
stations, and performance seems like it might be getting bad again, this time
probably due to reflections.
Back to the Farallon book, after reading through it, it seemed as terminating
a line to absord reflections was a definite must. Whether the wiring was a
daisy chain, bus, or active or passive star, they always stressed using the
120ohm TR. I accepted this as being realistic. What I needed to do was
figure out what resistance to use, if I wanted to put a TR at the end of
every branch (remember, I'm looking at 7-10 branches, more than double what
they recommend).
It turns out that if you follow their guidelines exactly, you will always
have at least 2 but never more than 4 TR's in the circuit. Since everything
on AppleTalk is in parallel, and the resistors as 120 ohms, that equates to
between 60 ohms (for 2 TR's) and 30 ohms (for 4 TR's) impedance across the
line, with closer to 60 ohms probably being better.
What I am thinking about doing is expanding this thinking to 8 or 10
branches, or TR's. Instead of using 120 ohm resistors at two ends, would
there be a problem with using 600 ohm resistors at each of my 10 branches?
Actually, what I'm trying to find out as if it seems correct to just balance
the resistance of the TR depending on how many branches there were, with
the object being to keep the overall impedance around 60 ohms or so? Right
now I have 6 branches, and since I can forsee adding probably 3 in the short
future, I'm leaning towards 470 ohm TR's. That would give my circuit an
impedance of about 78 ohms now (n=6) and about 52 ohms later (n=9), both pretty
close to the ideal of 60 ohms.
Assuming this theory works, would it be better to be above or below that
ideal value? For example, it I switched to 390 ohm TR's, at n=6 my impedance
would be 65 ohms, and at n=9 it would be 43 ohms. Which range would be
better? Also, would it be better to keep all the TR's the same, or fine
tune as I go along? For example, if I start out with 5 nodes with 300 ohm TR's
(net imp=60 ohms), if I add a 6th node, should I use a larger TR, so the overall
impedance doesn't fall as rapidly?
If anyone has done alything like this, I would appreciate hearing from you.
Unless I hear that it WON'T work, I will probably go ahead and try it.
Jeff White
Univ of Penn - CETS
jeff@eniac.seas.upenn.eduwnn@dsunx1.dsrd.ornl.gov (W. N. Naegeli) (04/11/89)
Jeff White ponders whether it would be beneficial to terminate each branch of a phone net with a resistor tuned to achieve a network impedance of 60 ohms. I don't have an anser to this problem, but it seems to me that this is exactly the theory between Nuvotechs TurboNet ST (self-terminating) connectors. Nuvotech claims that their ST connectors "add just the right amount" of resistance and "adjust this dynamically" as other nodes are added or taken off the network. We have an extensive phone net of very complex topology (multiple stars, some with small trees and daisy chains attached). In addition to 120 ohm resistors at the extreme ends of the trunk that connects the stars we are using TurboNet STs at most of the longer branches and for all noisy devices such as laser printers. Our current effective network length (counting branch lenght twice and 26-gauge flat extension cables four times) is about 4000'. There are 46 drops and 18 devices connected at present. As one would expect with such a crazy topology we have been experiencing some problems for the last few days since the last 12 drops were installed (some nodes can't see all other nodes). However, I expect these problems to vanish when the TurboBridge arrives that we ordered in anticipation of this situation and which will replace the Tops Repeater that is located at the middle of the trunk. In the meantime I can give preferrential treatment to some nodes by moving a couple of 120 ohm resistors among the ends of various branches. This seems to indicate to me that signal strength and network impedance, which stay the same are not as critical in our case as reflections and noise. That's why I expect the bridge to take care of the problems. The repeater also amplifies noise. I have not been very systematic in my termination trials, but my gut feeling is that the self-terminating connectors are beneficial in general, but that placing a 120 ohm resistor and a standard (non-terminating) connector at the ends of the trunk works better than using TurboNet STs at these locations. Another indication that noise is a principal cause of the problems is the fact that it makes a difference whether certain devices are on or off. Even when thay are idle, they seem to put out some noise. We need to be particularly careful that nobody switches the NEC SilentWriter LC-890 to parallel mode without first disconnecting it from the network, otherwise it brings down half of the network. NuvoTech does not reveal what they are doing inside their ST connectors and within what range it adds resistance. I'd be interested to find out. Has anyone taken an ST apart? Wolfgang N. Naegeli Oak Ridge National Laboratory wnn@dsunx1.dsrd.ornl.gov (128.219.96.46) (615) 574-6143
dheap@gara.une.oz (Dave Heap PSYS) (04/14/89)
In article <9750@netnews.upenn.edu> jeff@eniac.seas.upenn.edu (Jeffrey M White) writes: > > I'm in the process of setting up a larger than recommended passive star >7-10 branches instead of 3-4) using Phonenet type connectors and wiring, and >need some help with the final installation. (for those who may suggest it, >I know about Farallon's Star Controller, but the price is too high (> $1000), >and this installation is sort of a test installation anyway). > My problem has to do with the terminating resistors. For a long time, I >was confused as to whether the terminating resistors were: > 1. To absord reflections - in which case you wanted one at the end of every > line, or > 2. To keep a constant impedance across the line - in which case you would > only want several of them. The answer is (1), to absorb reflections. They work best when matched to the characteristic impedance of the line. Whether twisted telephone cable is really a 120ohm transmission line at the frequencies we are talking about is somewhat doubtful, however. (For an explanation of characteristic impedance & termination practices look up a textbook on transmission line theory - too involved to explain satisfactorily here). However, once you start adding branches you have problems. They should be terminated to reduce reflections, but each branch then increases loading on the system. (Even unterminated branches contribute to loading.) Farallon's recommendation of terminating the four longest branches is a compromise between reflection elimination & excessive load ( & therefore ability to drive the net). You'll find that they limit the number of recommended nodes on this sort of topology because of the drive problem (each node also loads the line). In summary, it's all a matter of compromise & whether you gain by putting more higher impedance (& hence less effective terminators) or less correct terminators is best determined empirically - if it works, fine. > It turns out that if you follow their guidelines exactly, you will always >have at least 2 but never more than 4 TR's in the circuit. Since everything >on AppleTalk is in parallel, and the resistors as 120 ohms, that equates to >between 60 ohms (for 2 TR's) and 30 ohms (for 4 TR's) impedance across the >line, with closer to 60 ohms probably being better. Not quite true, see above. > If anyone has done alything like this, I would appreciate hearing from you. >Unless I hear that it WON'T work, I will probably go ahead and try it. We've set up a passive star in excess of Farallon's recommendations & got away with it. It's a case of try it & see, keeping in mind the compromises you are making. -- Dave Heap ACSNET: dheap@gara.une.oz Psychology Department, UUCP: ...!uunet!munnari!gara.une.oz!dheap University of New England, ARPA: dheap%gara.une.oz@uunet.uu.net Armidale NSW 2351, Australia