Multi-Source Coded Downloads.

In this paper, we propose a selective-repeat (SR) automatic repeat-request (ARQ) model for multi-source download scenarios and analyze their useful throughput that we refer to as goodput. The multi-source scenario comprises a set of transmitters that send packets to a receiver. We characterize the forward channels from the transmitters to the receiver via a general hidden Markov model (HMM) and assume that the reverse channels from the receiver to the transmitter are lossless. To find the average goodput of the network, we exploit the probability-generation function. We consider different packet transmission schemes, including uncoded random, network coded and sliding window-based network coded packets, and contrast their performance. Our calculations show that using network coding in a multi-source scenario can increase the average goodput, while sliding window-based coding may also archive the theoretical maximum goodput. We show that our multi-source approach avoids the straggler problem, therefore adding more transmitters to the network increases its throughout and the system does not get limited by the weakest transmitter. We also verify our analytic results with extensive simulations.