Redundancy Techniques for Straggler Mitigation in Distributed Optimization and Learning
This addresses performance bottlenecks in distributed systems for machine learning, offering a practical solution with broad applicability, though it builds on existing redundancy concepts.
The paper tackles the problem of straggler nodes and slow communication links in distributed optimization and learning by proposing a framework that encodes datasets with redundancy, allowing dynamic exclusion of stragglers while maintaining convergence. It demonstrates deterministic convergence for various algorithms and shows performance improvements over uncoded, asynchronous, and data replication strategies in experiments on Amazon EC2 clusters.
Performance of distributed optimization and learning systems is bottlenecked by "straggler" nodes and slow communication links, which significantly delay computation. We propose a distributed optimization framework where the dataset is "encoded" to have an over-complete representation with built-in redundancy, and the straggling nodes in the system are dynamically left out of the computation at every iteration, whose loss is compensated by the embedded redundancy. We show that oblivious application of several popular optimization algorithms on encoded data, including gradient descent, L-BFGS, proximal gradient under data parallelism, and coordinate descent under model parallelism, converge to either approximate or exact solutions of the original problem when stragglers are treated as erasures. These convergence results are deterministic, i.e., they establish sample path convergence for arbitrary sequences of delay patterns or distributions on the nodes, and are independent of the tail behavior of the delay distribution. We demonstrate that equiangular tight frames have desirable properties as encoding matrices, and propose efficient mechanisms for encoding large-scale data. We implement the proposed technique on Amazon EC2 clusters, and demonstrate its performance over several learning problems, including matrix factorization, LASSO, ridge regression and logistic regression, and compare the proposed method with uncoded, asynchronous, and data replication strategies.