Thicknet/10base5 Test Segment: The Cable is In!

Paul Koning paulkoning at comcast.net
Wed Jun 27 08:18:41 CDT 2018



> On Jun 26, 2018, at 10:13 PM, Jon Elson via cctalk <cctalk at classiccmp.org> wrote:
> 
> On 06/26/2018 06:20 PM, Eric Smith via cctalk wrote:
>> On 06/26/2018 03:15 PM, Grant Taylor via cctalk wrote:
>>>> I can only guess that having a terminator too close interferes with or
>>>> weakens the signal too much in some way.
>>> 
> No, I think it may have something to do with properly detecting all collisions.  There are a whole bunch of special cases, where short packets have crossed in the middle of a segment.  This causes a collision at the nodes in the center of the segment, but the nodes at the ends see their own transmissions without interference. Possibly, having the terminator too close to (one of) the sending nodes might make this detection less reliable.   Hmmm, but really, anything that goes past the last tap toward the terminator ought to just DISAPPEAR, so that the length beyond the tap should not matter.

It's clear to me: the person making the claim to Jonathan simply didn't know enough EE to understand that his statement was invalid.  No, the terminator placement has nothing to do with it.

Collision detect is a very simple process: it involves measuring the signal on the wire and comparing that with what the transceiver is currently sending.  If the two differ by enough, it means another sender is active at the same time.  There are no special cases in this.

Proper collision detection depends on the segment length rule and the minimum packet length.  If you run the numbers, you will see that the minimum packet length and max segment length were chosen to ensure that a collision between two stations, at opposite end of a max length cable, will be detected by both senders.  If you violate either limit by a significant amount this will no longer be true, and you've reduced the system to basically Aloha, which works fine at low load but maxes out at 16% of wire rate, or thereabouts.

	paul



More information about the cctech mailing list