@inproceedings{c583ec4c3693435394ce37fa30d1ba98,
title = "In-situ workflow auto-tuning through combining component models",
abstract = "In-situ parallel workflows couple multiple component applications via streaming data transfer to avoid data exchange via shared file systems. Such workflows are challenging to configure for optimal performance due to the huge space of possible configurations. Here, we propose an in-situ workflow auto-tuning method, ALIC, which integrates machine learning techniques with knowledge of in-situ workflow structures to enable automated workflow configuration with a limited number of performance measurements. Experiments with real applications show that ALIC identify better configurations than existing methods given a computer time budget.",
keywords = "auto-tuning, component model, in-situ workflow",
author = "Tong Shu and Yanfei Guo and Justin Wozniak and Xiaoning Ding and Ian Foster and Tahsin Kurc",
note = "Funding Information: We measured the actual computer time of the best configurations of LV and GP predicted by RS, GEIST, AL, and ALIC, and plot the measurements for of 25 and 50 in Fig. 1, which shows that the computer times achieved by ALIC are always better than by RS, GEIST, and AL. ALIC outperforms AL, because surrogate models trained with the same number of training samples are much more accurate for component applications than in-situ workflows, and our method of estimating workflow performance provides relatively accurate configuration ranking over top configurations. Acknowledgments This research was supported by the Exascale Computing Project (17-SC-20-SC) from U.S. Department of Energy and start-up funds from Southern Illinois University Carbondale. References Publisher Copyright: {\textcopyright} 2021 Owner/Author.; 26th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP 2021 ; Conference date: 27-02-2021 Through 03-03-2021",
year = "2021",
month = feb,
day = "17",
doi = "10.1145/3437801.3441615",
language = "English (US)",
series = "Proceedings of the ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPOPP",
publisher = "Association for Computing Machinery",
pages = "467--468",
booktitle = "PPoPP 2021 - Proceedings of the 2021 26th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming",
}