The mining of multi-dimensional time series is a crucial step in gaining insights into data obtained from physical systems and from monitoring infrastructures. A widely accepted approach for this challenge is the matr...
详细信息
ISBN:
(纸本)9781665481069
The mining of multi-dimensional time series is a crucial step in gaining insights into data obtained from physical systems and from monitoring infrastructures. A widely accepted approach for this challenge is the matrix profile, which, however, is computationally very expensive. It relies on calculating large correlation matrices coupled with sort operations across all dimensions of the data, as well as on performing inclusive scans. All of these steps are inherently data parallel and can, therefore, benefit from execution on gpus, and even more so from horizontal scaling on multiple gpus. In addition, the nature of the matrix profile calculation allows the exploitation of reduced precision on gpus. This offers further improvements to enable the analysis of ever growing data sets in real-world scenarios. Based on these motivations, we introduce the first parallel algorithm for multi-dimensional matrix profile on multiple gpus exploiting reduced precision modes and provide a highly optimized implementation using novel optimization techniques. On one NVIDIA A100 gpu, our implementation achieves a 54x performance improvement in comparison to an optimized singlenode execution on a state-of-the-art CPU-based implementation relying on double-precision computation and an additional factor of 1.4x when switching to reduced precision while maintaining sufficient accuracy. We study the accuracy and performance trade-offs for our proposed algorithm in detail and present synthetic and real-world case studies to demonstrate how the reduced precision improves the performance, while accomplishing sufficiently accurate results.
Weather Research Forecasting (WRF) model is a popular tool used in both research and operational applications that enables us to obtain relevant information about rain, snow, and others. The WRF Single-Moment 6-Class ...
详细信息
ISBN:
(纸本)9781665410168
Weather Research Forecasting (WRF) model is a popular tool used in both research and operational applications that enables us to obtain relevant information about rain, snow, and others. The WRF Single-Moment 6-Class Microphysics Scheme (WSM6) is an important routine and it is the most lime-consuming task of the WRF model. Due to its importance, several parallel methods have been proposed to the problem. The parallel approach using multiple gpu for the WSM6 scheme has allowed the address of a large number of data in a reasonable time. This paper describes the improvement of the computational performance of the WSM6 by exploiting fine-grained parallelism using the Graphics Processing Unit (gpu) with OpenACC paradigm. When compared to a 24-thread CPU, the speedup we achieve is 67.4 and 108.0 on one and four gpus, respectively. We also compare our implementation to a recent OpenACC implementation in the literature and our proposed solution is 21 times faster than it using one gpu and 34 times faster using four gpus. We also performed a study about the accuracy of the proposed implementation with good results.
multiple sequence alignment is an important tool to represent similarities among biological sequences and it allows obtaining relevant information such as evolutionary history, among others. Due to its importance, sev...
详细信息
ISBN:
(纸本)9783030866532;9783030866525
multiple sequence alignment is an important tool to represent similarities among biological sequences and it allows obtaining relevant information such as evolutionary history, among others. Due to its importance, several methods have been proposed to the problem. However, the inherent complexity of the problem allows only non-exact solutions and further for small length sequences or few sequences. Hence, the scenario of rapid increment of the sequence databases leads to prohibitive runtimes for large-scale sequence datasets. In this work we describe a multi-gpu approach for the three stages of the Progressive Alignment method which allow to address a large number of lengthy sequence alignments in reasonable time. We compare our results with two popular aligners ClustalW-MPI and Clustal Omega and with CUDA NW module of the Rodinia Suite. Our proposal with 8 gpus achieved speedups ranging from 28.5 to 282.6 with regard to ClustalW-MPI with 32 CPUs considering NCBI and synthetic datasets. When compared to Clustal Omega with 32 CPUs for NCBI and synthetic datasets we had speedups between 3.3 and 32. In comparison with CUDA NW_Rodinia the speedups range from 155 to 830 considering all scenarios.
暂无评论