Expediting Distributed DNN Training with Device Topology-Aware Graph Deployment

TL;DR

This paper introduces TAG, a system that optimizes distributed DNN training by considering device topology and computation graphs, achieving significant speed-ups and adaptable deployment strategies across heterogeneous clusters.

Contribution

The paper proposes a novel GNN-based approach combined with search methods to optimize DNN training graphs for heterogeneous device topologies, including a lossless gradient compression technique.

Findings

Achieves up to 4.56x training speed-up

Effective for unseen models and topologies

Reduces communication overhead

Abstract

This paper presents TAG, an automatic system to derive optimized DNN training graph and its deployment onto any device topology, for expedited training in device- and topology- heterogeneous ML clusters. We novelly combine both the DNN computation graph and the device topology graph as input to a graph neural network (GNN), and join the GNN with a search-based method to quickly identify optimized distributed training strategies. To reduce communication in a heterogeneous cluster, we further explore a lossless gradient compression technique and solve a combinatorial optimization problem to automatically apply the technique for training time minimization. We evaluate TAG with various representative DNN models and device topologies, showing that it can achieve up to 4.56x training speed-up as compared to existing schemes. TAG can produce efficient deployment strategies for both unseen DNN models and unseen device topologies, without heavy fine-tuning.