Matroid theory based broadcast retransmission in distributed storage network

Abstract

In conventional distributed storage networks, backbone firstly sends the packets that needed by the users randomly to existed servers. Then backbone broadcasts the packets uncoded to the users. This phase is defined as systematic transmission phase. Due to channel distortion and noise, different users may lost different packets. Thus, each user needs to feedback its packet loss state and could constitute a state feedback matrix (SFM) in the side of the backbone. After this, instead of retransmission by the backbone, the binary network coded or original packets are sent from servers to the users, and each user can only receive a packet from one server within a time slot. In this paper, we consider backbone network only sends the packets to the servers after receiving the SFM from all users, and we assume that the time needed for the transmissions from backbone to the servers can be ignored. According to the matroid theory, we propose an algorithm that decide the set of packets that needed to send to the different servers in order to decrease the server resource consumption and utilize the servers to further decrease the number of retransmissions.