Fast Resource Allocation for Network-Coded Traffic - A Coded-Feedback Approach

In this paper, we develop a fast resource allocation algorithm that takes advantage of intra-session network coding. The algorithm maximizes the total utility of multiple unicast (or multicast) sessions subject to capacity constraints, where packets are coded within each session. Our solution is a primal solution that does not use duality or congestion prices. Thus, it does not require building up queues to achieve the optimal resource allocation. Hence, the queueing delay of the packets can be tightly controlled. The existing primal solution in the literature requires a separate graph-theoretic algorithm to find the min-cut of each session, whose complexity grows quadratically with the total number of nodes. In contrast, we provide a new coded-feedback approach whose complexity grows only linearly with the total number of nodes. More explicitly, by letting the ACK/feedback packets on the return paths also carry coding coefficients as does the forward coded traffic, key network information can be obtained more efficiently, which leads to a fast resource allocation scheme fully integrated with the network coding operation.