On the latency of heterogeneous MDS queue

Multi-tenant distributed storage systems (DSS) exhibit heterogeneity in several dimensions such as fault tolerance requirements, nature of job requests etc. However the primary focus in literature has been on homogenous models for DSS. In this paper, we investigate the impact of heterogeneity on the latency performance of multi-tenant distributed storage systems. Heterogeneity across multiple tenants is modeled via potentially different fault tolerance requirements and different job arrival rates. We consider a heterogeneous multi-tenant DSS with n servers that store the data of R distinct traffic classes. The data of each traffic class i is stored in the DSS using a different (n, fc¿) Maximum-Distance-Separable (MDS) code. Depending upon the traffic class, the data may be frequently or infrequently accessed and is modeled using different job arrival rates for the traffic classes. We then present a queuing theoretic analysis of the proposed model and establish upper and lower bounds on the average latency for each traffic class for various scheduling policies. Using simulations, we verify the accuracy of the derived bounds and present qualitative insights on the impact of heterogeneity and scheduling policies on the mean latency of different classes.