# Hop: Heterogeneity-Aware Decentralized Training

## TL;DR

Hop is a novel heterogeneity-aware decentralized training protocol that improves performance in heterogeneous environments by using a queue-based synchronization mechanism and iteration skipping, demonstrated with significant speedups on CNN and SVM.

## Contribution

This paper introduces Hop, the first decentralized training protocol designed to handle heterogeneity effectively, using a queue-based synchronization and iteration skipping.

## Key findings

- Significant speedup over standard decentralized training in heterogeneous settings.
- Effective mitigation of slow workers through iteration skipping.
- Prototype implementation on TensorFlow demonstrates practical viability.

## Abstract

Recent work has shown that decentralized algorithms can deliver superior performance over centralized ones in the context of machine learning. The two approaches, with the main difference residing in their distinct communication patterns, are both susceptible to performance degradation in heterogeneous environments. Although vigorous efforts have been devoted to supporting centralized algorithms against heterogeneity, little has been explored in decentralized algorithms regarding this problem. This paper proposes Hop, the first heterogeneity-aware decentralized training protocol. Based on a unique characteristic of decentralized training that we have identified, the iteration gap, we propose a queue-based synchronization mechanism that can efficiently implement backup workers and bounded staleness in the decentralized setting. To cope with deterministic slowdown, we propose skipping iterations so that the effect of slower workers is further mitigated. We build a prototype implementation of Hop on TensorFlow. The experiment results on CNN and SVM show significant speedup over standard decentralized training in heterogeneous settings.