Saved in:
Bibliographic Details
Main Authors: Tan, Qiao, Zhu, Feng, Zhang, Jingjing
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2301.08895
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929215013650432
author Tan, Qiao
Zhu, Feng
Zhang, Jingjing
author_facet Tan, Qiao
Zhu, Feng
Zhang, Jingjing
contents Wall-clock convergence time and communication rounds are critical performance metrics in distributed learning with parameter-server setting. While synchronous methods converge fast but are not robust to stragglers; and asynchronous ones can reduce the wall-clock time per round but suffers from degraded convergence rate due to the staleness of gradients, it is natural to combine the two methods to achieve a balance. In this work, we develop a novel asynchronous strategy that leverages the advantages of both synchronous methods and asynchronous ones, named adaptive bounded staleness (ABS). The key enablers of ABS are two-fold. First, the number of workers that the PS waits for per round for gradient aggregation is adaptively selected to strike a straggling-staleness balance. Second, the workers with relatively high staleness are required to start a new round of computation to alleviate the negative effect of staleness. Simulation results are provided to demonstrate the superiority of ABS over state-of-the-art schemes in terms of wall-clock time and communication rounds.
format Preprint
id arxiv_https___arxiv_org_abs_2301_08895
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle ABS: Adaptive Bounded Staleness Converges Faster and Communicates Less
Tan, Qiao
Zhu, Feng
Zhang, Jingjing
Distributed, Parallel, and Cluster Computing
Wall-clock convergence time and communication rounds are critical performance metrics in distributed learning with parameter-server setting. While synchronous methods converge fast but are not robust to stragglers; and asynchronous ones can reduce the wall-clock time per round but suffers from degraded convergence rate due to the staleness of gradients, it is natural to combine the two methods to achieve a balance. In this work, we develop a novel asynchronous strategy that leverages the advantages of both synchronous methods and asynchronous ones, named adaptive bounded staleness (ABS). The key enablers of ABS are two-fold. First, the number of workers that the PS waits for per round for gradient aggregation is adaptively selected to strike a straggling-staleness balance. Second, the workers with relatively high staleness are required to start a new round of computation to alleviate the negative effect of staleness. Simulation results are provided to demonstrate the superiority of ABS over state-of-the-art schemes in terms of wall-clock time and communication rounds.
title ABS: Adaptive Bounded Staleness Converges Faster and Communicates Less
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2301.08895