Implementing Distributed Storage Systemsby Network Coding and ConsideringComplexity of Decoding

Sammanfattning: Recently, network coding for distributed storage system has become a popular field due to increasing applications such as video, VoIP or mail. There are lots of theoretical works in this field, yet not enough practical study. In this thesis we implement a distributed storage system using network coding. In our implementation, three strategies of coding applied to this system: replication, regenerating code and regenerating code with repair by transfer. To study advantageous or disadvantageous of these strategies, we measure probability of successful downloading, repair time and processing time after implementation. We further study regenerating code with different finite field. Moreover we propose a method for low complexity of decoding algorithm. It is to assign different number of connected storage node which a receiver uses to reconstruct an original file. Our results show that the regenerating code with repair by transfer is an optimal network code for the distributed storage system when comparing to other strategies when working in small finite field size. In particular, in GF(2), the code only uses exclusive-OR to encode and decode data. In addition when finite field is large, the probability of successful downloading increases with the cost of higher complexity comparing to network code with small finite field size. To work in small finite field and consequently reducing complexity in decoding, we show by increasing number of connected node the probability of successful downloading improves. Thus we conclude that the regenerating code with repair by transfer is optimal implementation within system. However if we only consider the regenerating code with different number of connected storage node retrieving the original file, higher number of connected storage node is better than lower number of storage node connected.