pgc@newcastle.ac.uk (P. G. Cutting) (05/06/88)
What are the rules governing segments and repeaters? Are there repeaters which allow topologies other then linear and how do the rules apply in this case? What is the main factor that the rules seek to control in this context? I suspect end-end prop delay since this will affect the collisions and hence channel bandwidth. If the end-end prop delay becomes too large will it cause catestrophic failure (eg > 9.6us) or is the degredation continuous.
eshop@saturn.ucsc.edu (Jim Warner) (05/08/88)
In article <3105@cheviot.newcastle.ac.uk> pgc@newcastle.ac.uk (P. G. Cutting) writes: >What are the rules governing segments and repeaters? For IEEE 802.3 repeaters the rule is that the no pair of stations can be separated by more than four repeaters. Of the three segments separating the segments containing these stations, two must be link segments. A link segment is one with no transceivers except for those that attach the repeaters. It is permissible for the two link segments to be a double length (1 km) fiber segment. The rules are different for repeaters designed to the Ethernet version 2 specification. Caveate. >Are there repeaters which allow topologies other then linear >and how do the rules apply in this case? This is allowed for all repeaters. The rule is "max # of repeaters between most distant stations." Note that if you use an n-way repeater to create an n-way star, passing through it counts as "one." >What is the main factor that the rules seek to control >in this context? I suspect end-end prop delay since this >will affect the collisions and hence channel bandwidth. It is round trip propagation delay; carrier sense in the forward direction and collision sense in the reverse direction. The objective is that collision fragments should be short enough that they can be thrown out on the basis of length. Legal collision fragments are smaller than a minimum packet. >If the end-end prop delay becomes too large will it cause >catestrophic failure (eg > 9.6us) or is the degredation >continuous. If the network is so large that a node at one extreme can send a short packet and not be informed that a collision has occured while it is still transmitting, it will not know that this packet was garbled at its intended destination. Upper software levels will have to detect something amiss and retransmit. For well designed protocols the degradation should be continuous and gradual. 9.6 uS is the minimum interpacket spacing in 802.3. The max round trip collision propagation time is 512 bits = 51 uS.