Title:	TIGHTLY COUPLED IN SITU VISUALIZATION WITH SIMULATIONS ON THOUSANDS OF CPU CORES
Author(s):	Zhiwei Ai, Yi Cao, Li Xiao, Huawei Wang and Aiqing Zhang
ISBN:	978-989-8533-66-1
Editors:	Yingcai Xiao and Ajith P. Abraham
Year:	2017
Edition:	Single
Keywords:	In situ visualization, extrascale simulation, JASMIN infrastructure, common component architecture
Type:	Full Paper
First Page:	155
Last Page:	162
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Scientific visualization is an effective way to analyze the complex datasets generated from numerical simulations. Current supercomputer I/O performance struggles to match the rapidly improving computing performance. Therefore, there is in urgent need for in situ visualization to reduce the amount of data output during large-scale scientific simulations. This paper presents an in situ component technique that enables a general purpose visualization tool to be flexibly coupled with the simulations based on JASMIN, a parallel programming framework for large-scale numerical simulations. Using the presented technique, the data analysis and visualization tasks are shifted to run in situ, that is, closer to the simulation via resource sharing. It is beneficial to avoid storing large-scale datasets for post-processing. Subsequently, a tightly coupled in situ visualization mode is proposed, in which visualization code can be deployed into application code with low code-integration cost. This code can then run on the same supercomputer resources as the simulation code. The JASMIN-based in situ component provides a bridge between a batch of parallel simulation applications and existing visualization packages. The implemented component allows applications to integrate the rich functionality of existing visualization tools with a unified data-management interface. It also enables efficient use of supercomputer resources through a batch processing mechanism. In situ visualization of the electromagnetic interference surrounding an aircraft is simulated using a model that comprises 8.5 billion grid cells on the TianHe-2 supercomputer.
	Go Back