Sky Computing: Accelerating Geo-distributed Computing in Federated Learning

TL;DR

Sky Computing introduces a load-balanced model parallelism framework for federated learning, optimizing weight allocation based on device capabilities, significantly reducing training time for large models across distributed devices.

Contribution

The paper proposes Sky Computing, a novel load-balanced model parallelism framework that adaptively allocates model weights according to device capabilities in federated learning.

Findings

Outperforms baseline by 55% in training time for 160-layer BERT

Effectively balances load across heterogeneous devices

Reduces training time significantly in federated settings

Abstract

Federated learning is proposed by Google to safeguard data privacy through training models locally on users' devices. However, with deep learning models growing in size to achieve better results, it becomes increasingly difficult to accommodate the whole model on one single device. Thus, model parallelism is then used to divide the model weights among several devices. With this logic, the approach currently used evenly allocates weights among devices. However, in reality, a computation bottleneck may occur resulting from variant computing power of different users' devices. To address this problem, load balancing is needed to allocate the model weights based on the computational capability of the device. In this paper, we proposed Sky Computing, a load-balanced model parallelism framework to adaptively allocate the weights to devices. Sky Computing outperforms the baseline method by 55% in training time when training 160-layer BERT with 64 nodes. The source code can be found at https://github.com/hpcaitech/SkyComputing.