On a class of Distributed Storage Systems with Prioritized Data Sources
Congduan Li · 2019
The rate regions and sufficiency of simple linear codes for a class of distributed storage systems with prioritized data sources are investigated in this paper. The exact repair problem is considered. Different from conventional setups, the scenario considered in this paper has prioritized data sources, where the hot data has higher priority than cold data in the decoding process. It is assumed that a user demanding cold data demands hot data as well. Instead of using same capacity for all storage nodes, the rate regions of interest are all feasible different storage sizes versus different tuples of source entropies, with assumption of sufficient large repair bandwidth. Both symmetric and asymmetric repairs are discussed in the paper. Linear network codes over some finite field are said to be sufficient for such a distributed storage system if and only if for every point in the rate region, there exists a code over that finite field to achieve it. As a multi-source multi-sink network coding problem, the rate regions are obtained from computer-aided approaches via bounding the region of entropic vectors. Experimental results on the rate regions of hundreds of non-isomorphic distributed storage systems are presented for demonstration. In addition, it is shown that binary linear codes suffice for most of them.