A Time-Series Model for Varying Worker Ability in Heterogeneous Distributed Computing Systems

In this paper, we consider the problem of estimating time-dependent ability workers participating in distributed matrix-vector multiplication over heterogeneous clusters. Specifically, model workers’ as a latent variable and introduce log-normally working rate function with parameters so that increases increases, takes positive values only. This modeling is motivated by need to reflect impact external factors on performance. We estimate using expectation-maximization (EM) algorithm combined particle method. The proposed estimation inference rates are used allocate tasks reduce expected latency. From simulations, observe our effective reducing