Existing Data Parallel (DP) trainings for deep neural networks (DNNs) often experience limited scalability in speedup due to substantial communication overheads. While Overlapping technique can mitigate such problem by paralleling communication and computation in DP, its effectiveness is constrained by the high communication-to-computation ratios (CCR) of DP training tasks. Gradient compression (GC) is a promising technique to obtain lower CCR by reducing communication volume directly. However, it is challenging to obtain real performance improvement by applying GC into Overlapping because of (1) severe performance penalties in traditional GCs caused by high compression overhead and (2) decline of Overlapping benefit owing to the possible data dependency in GC schemes. In this paper, we propose COVAP, a novel GC scheme designing a new coarse-grained filter, makes the compression overhead close to zero. COVAP ensures an almost complete overlap of communication and computation by employing adaptive compression ratios and tensor sharding tailored to specific training tasks. COVAP also adopts an improved error feedback mechanism to maintain training accuracy. Experiments are conducted on Alibaba Cloud ECS instances with different DNNs of real-world applications. The results illustrate that COVAP outperforms existent GC schemes in time-to-solution by 1.92x-15.39x and exhibits near-linear scaling. Furthermore, COVAP achieves best scalability under experiments on four different cluster sizes.
Lossy compression is one of the most efficient solutions to reduce storage overhead and improve I/O performance for HPC applications. However, existing parallel I/O libraries cannot fully utilize lossy compression to accelerate parallel write due to the lack of deep understanding on compression-write performance. To this end, we propose to deeply integrate predictive lossy compression with HDF5 to significantly improve the parallel-write performance. Specifically, we propose analytical models to predict the time of compression and parallel write before the actual compression to enable compress…
The performance and efficiency of distributed training of Deep Neural Networks highly depend on the performance of gradient averaging among all participating nodes, which is bounded by the communication between nodes. There are two major strategies to reduce communication overhead: one is to hide communication by overlapping it with computation, and the other is to reduce message sizes. The first solution works well for linear neural architectures, but latest networks such as ResNet and Inception offer limited opportunity for this overlapping. Therefore, researchers have paid more attention to…
Large scale simulations of complex systems ranging from climate and astrophysics to crowd dynamics, produce routinely petabytes of data and are projected to reach the zettabytes level in the coming decade. These simulations enable unprecedented insights but at the same their effectiveness is hindered by the enormous data sizes associated with the computational elements and respective output quantities of interest that impose severe constraints on storage and I/O time. In this work, we address these challenges through a novel software framework for scientific data compression. The software (Cub…
Yonsei University（延世大学）；Seoul National University（首尔国立大学）；Samsung Electronics（三星电子）
In training of modern large natural language processing (NLP) models, it has become a common practice to split models using 3D parallelism to multiple GPUs. Such technique, however, suffers from a high overhead of inter-node communication. Compressing the communication is one way to mitigate the overhead by reducing the inter-node traffic volume; however, the existing compression techniques have critical limitations to be applied for NLP models with 3D parallelism in that 1) only the data parallelism traffic is targeted, and 2) the existing compression schemes already harm the model quality to…
TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training
Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences（中国科学院大学杭州高等研究院）；Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）；University of Science and Technology of China（中国科学技术大学）
Modern HPC applications produce increasingly large amounts of data, which limits the performance of current extreme-scale systems. Data reduction techniques, such as lossy compression, help to mitigate this issue by decreasing the size of data generated by these applications. SZ, a current state-of-the-art lossy compressor, is able to achieve high compression ratios, but the prediction/quantization methods used introduce dependencies which prevent parallelizing this step of the compression. Recent work proposes a parallel dual prediction/quantization algorithm for GPUs which removes these depe…
As HPC systems continue to grow to exascale, the amount of data that needs to be saved or transmitted is exploding. To this end, many previous works have studied using error-bounded lossy compressors to reduce the data size and improve the I/O performance. However, little work has been done for effectively offloading lossy compression onto FPGA-based SmartNICs to reduce the compression overhead. In this paper, we propose a hardware-algorithm codesign for an efficient and adaptive lossy compressor for scientific data on FPGAs (called CEAZ), which is the first lossy compressor that can achieve h…
A rich body of prior work has highlighted the existence of communication bottlenecks in synchronous data-parallel training. To alleviate these bottlenecks, a long line of recent work proposes gradient and model compression methods. In this work, we evaluate the efficacy of gradient compression methods and compare their scalability with optimized implementations of synchronous data-parallel SGD across more than 200 different setups. Surprisingly, we observe that only in 6 cases out of more than 200, gradient compression methods provide speedup over optimized synchronous data-parallel training i…
University of Rochester（罗切斯特大学）；Hong Kong University of Science and Technology（香港科技大学）；Kwai Inc（快手公司）
A standard approach in large scale machine learning is distributed stochastic gradient training, which requires the computation of aggregated stochastic gradients over multiple nodes on a network. Communication is a major bottleneck in such applications, and in recent years, compressed stochastic gradient methods such as QSGD (quantized SGD) and sparse SGD have been proposed to reduce communication. It was also shown that error compensation can be combined with compression to achieve better convergence in a scheme that each node compresses its local stochastic gradient and broadcast the result…
Decompressing a file made by the gzip program at an arbitrary location is in principle impossible, due to the nature of the DEFLATE compression algorithm. Consequently, no existing program can take advantage of parallelism to rapidly decompress large gzip-compressed files. This is an unsatisfactory bottleneck, especially for the analysis of large sequencing data experiments. Here we propose a parallel algorithm and an implementation, pugz, that performs fast and exact decompression of any text file. We show that pugz is an order of magnitude faster than gunzip, and 5x faster than a highly-opti…
ETH Zürich（苏黎世联邦理工学院）；Stanford University（斯坦福大学）；Zhejiang University（浙江大学）；Carnegie Mellon University（卡内基梅隆大学）；Meta AI
Communication compression is a crucial technique for modern distributed learning systems to alleviate their communication bottlenecks over slower networks. Despite recent intensive studies of gradient compression for data parallel-style training, compressing the activations for models trained with pipeline parallelism is still an open problem. In this paper, we propose AC-SGD, a novel activation compression algorithm for communication-efficient pipeline parallelism training over slow networks. Different from previous efforts in activation compression, instead of compressing activation values d…
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.
We propose Slim Graph: the first programming model and framework for practical lossy graph compression that facilitates high-performance approximate graph processing, storage, and analytics. Slim Graph enables the developer to express numerous compression schemes using small and programmable compression kernels that can access and modify local parts of input graphs. Such kernels are executed in parallel by the underlying engine, isolating developers from complexities of parallel programming. Our kernels implement novel graph compression schemes that preserve numerous graph properties, for exam…
The automatic identification system (AIS), an automatic vessel-tracking system, has been widely adopted to perform intelligent traffic management and collision avoidance services in maritime Internet of Things (IoT) industries. With the rapid development of maritime transportation, tremendous numbers of AIS-based vessel trajectory data have been collected, which make trajectory data compression imperative and challenging. This paper mainly focuses on the compression and visualization of large-scale vessel trajectories and their Graphics Processing Unit (GPU)-accelerated implementations. The vi…
