The parallel use of multiple channels in a WDM star network means that too
many packets may simultaneously arrive for the same destination station, ne
cessitating the implementation of a destination-conflict-resolution functio
n somewhere within the network. This paper considers explicitly, the placem
ent of the destination-conflict-resolution function which specifies the loc
ation(s) where it should be performed, and when it should be performed. Tra
ditional placements in which the function is located at all user stations a
nd performed either before packet transmission (using the request-schedule-
then-transmit principle) or after a destination conflict has been detected
(using the detect-and-retransmit-if-lost principle), is compared with a cen
tral placement in which only one central station located at the entrance to
the star coupler is responsible for detecting conflicts and optically buff
ering 'otherwise lost' packets whilst they are en route to their destinatio
ns, until their destinations are free to receive them. The networks are eva
luated considering their delay and throughout characteristics, the computat
ional complexity of their protocols, their fault tolerance, and their hardw
are demands. The results presented suggest that significant performance imp
rovements are achievable with the central placement since destination confl
icts are resolved without having to retransmit packets nor waiting until th
e end of a request-broadcast-and-schedule phase before a given packet can b
e transmitted. The central station works with Space Division Multiplexed si
gnals, just before they enter the star coupler. Its implementation is there
fore simpler than when all stations are charged with this task, each of whi
ch has to attend to multiple WDM channels. Only 'otherwise lost' packets ar
e buffered so the network has low buffer memory requirements. (C) 1999 Else
vier Science B.V. All rights reserved.