Balancing Straggler Mitigation and Information Protection for Matrix Multiplication in Heterogeneous Multi-Group Networks

Abstract

Distributed computing has made it possible to satisfy the demands for large-scale matrix multiplication. A distributed computing system suffers from both straggler problem and information leakage. In a heterogeneous network consisting of multiple worker groups, information leakage can be caused by both intra-group and inter-group collusion. Besides, considering the heterogeneity of worker nodes, stronger nodes are supposed to compute more tasks to provide robustness for stragglers. However, this results in more information being leaked to stronger nodes, contradicting the principle of information protection. In this paper, we propose a multi-group heterogeneous secure coded matrix multiplication (MG-HSCMM) scheme to solve these problems in a heterogeneous multi-group network. By taking the heterogeneity of worker nodes into consideration, the corresponding recovery threshold and security constraint are obtained. To improve the performance of such a network, a low complexity task allocation policy that balances straggler mitigation and information protection is given. Compared with existing schemes, MG-HSCMM can achieve significant performance gain. Numerical simulation results verify the superiority of our proposed scheme.