CD-SGD: Distributed Stochastic Gradient Descent with Compression and Delay Compensation

Communication overhead is the key challenge for distributed training. Gradient compression is a widely used approach to reduce communication traffic. When combining with parallel communication mechanism method like pipeline, gradient compression technique can greatly alleviate the impact of communication overhead. However, there exists two problems of gradient compression technique to be solved. Firstly, gradient compression brings in extra computation cost, which will delay the next training iteration. Secondly, gradient compression usually leads to the decrease of convergence accuracy.

OD-SGD: One-step Delay Stochastic Gradient Descent for Distributed Training

The training of modern deep learning neural network calls for large amounts of computation, which is often provided by GPUs or other specific accelerators. To scale out to achieve faster training speed, two update algorithms are mainly applied in the distributed training process, i.e. the Synchronous SGD algorithm (SSGD) and Asynchronous SGD algorithm (ASGD). SSGD obtains good convergence point while the training speed is slowed down by the synchronous barrier. ASGD has faster training speed but the convergence point is lower when compared to SSGD. To sufficiently utilize the advantages of SSGD and ASGD, we propose a novel technology named One-step Delay SGD (OD-SGD) to combine their strengths in the training process. Therefore, we can achieve similar convergence point and training speed as SSGD and ASGD separately. To the best of our knowledge, we make the first attempt to combine the features of SSGD and ASGD to improve distributed training performance. Each iteration of OD-SGD contains a global update in the parameter server node and local updates in the worker nodes, the local update is introduced to update and compensate the delayed local weights. We evaluate our proposed algorithm on MNIST, CIFAR-10 and ImageNet datasets. Experimental results show that OD-SGD can obtain similar or even slightly better accuracy than SSGD, while its training speed is much faster, which even exceeds the training speed of ASGD.

Communication Optimization Strategies for Distributed Deep Learning: A Survey

Recent trends in high-performance computing and deep learning lead to a proliferation of studies on large-scale deep neural network (DNN) training. However, the frequent communication requirements among computation nodes drastically slow down the overall training speed, which makes the bottleneck in distributed training, particularly in clusters with limited network bandwidth. To mitigate the drawbacks of distributed communication, researchers have proposed various optimization strategies. In this paper, we give a comprehensive survey of communication strategies from both algorithm and computer network perspectives. Algorithm optimizations focus on reducing the amount of communication in distributed training, while network optimizations focus on speeding up the communication between distributed devices. At the algorithm level, we describe how to reduce the number of communication rounds and transmitted bits per round, besides we shed light on how to overlap computation and communication. At the network level, we discuss the effect caused by network infrastructures, including communication schemes, network protocols, and topology. Finally, we extrapolate potential challenges and research directions for communication acceleration in distributed DNN training.