Latency Analysis For Distributed Storage

Modern communication and computation systems consist of large networks of unreliable nodes. Yet, it is well known that such systems can provide aggregate reliability via information redundancy, duplicating paths, or replicating computations. While redundancy may increase the load on a system, it can also lead to major performance improvements through the judicious management of additional system resources. Two important examples of this abstract paradigm are content access from multiple caches in content delivery networks and master/slave computations on compute clusters. Many recent articles in the area have proposed bounds on the latency performance of redundant systems, characterizing the latency-redundancy tradeoff under specific load profiles. Following a similar line of research, this article introduces new analytical bounds and approximation techniques for the latency-redundancy tradeoff for a range of system loads and two popular redundancy schemes. The proposed framework allows for approximating the equilibrium latency distribution, from which various metrics can be derived including mean, variance, and the tail decay of stationary distributions.