Optimal Load Allocation for Coded Distributed Computation in Heterogeneous Clusters

Recently, coding has been a useful technique to mitigate stragglers' effect in distributed computing. However, this context mainly explored assuming homogeneous workers, although real-world clusters often consist of heterogeneous workers with different computing capabilities. The uniform load allocation without considering the heterogeneity possibly causes significant loss latency. In article, we suggest optimal for coded workers. Specifically, focus on scenario that there exist having same capability, which can be regarded as group analysis. We rely lower bound expected latency and obtain by showing our achieves minimum sufficiently large number Given proposed allocation, derive code rate achieve From numerical simulations, when heterogeneity, reduces orders magnitude over existing scheme. Furthermore, from experiments Amazon EC2 scenarios distinct straggler/heterogeneity patterns, observe scheme outperforms competing schemes reducing total finishing time up 52%.