We study the problem of dynamic routing on arrays. We prove that a lar
ge class of greedy algorithms perform very well on average. In the dyn
amic case, when the arrival rate of packets in an N x N array is at mo
st 99% of network capacity, we establish an exponential bound on the t
ail of the delay distribution. Moreover, we show that in any window of
T steps, the maximum queue-size is O(1 + log T/log N) with high proba
bility. We extend these results to the case of bit-serial routing, and
to the static case. We also calculate the exact value of the ergodic
expected delay and queue-sizes under the farthest first protocol for t
he one-dimensional array, and for the ring when the arrivals are Poiss
on. (C) 1998 Academic Press.