Scientific applications, simulations and large-scale experiments generate an ever increasing deluge of data. Due to the storage hardware not being able to keep pace with the amount of computational power, data reduction techniques have to be employed. Care has to be taken such that data reduction does not impact energy efficiency as it is an important cost factor for supercomputer systems and infrastructures. Data reduction techniques are highly data-specific and, therefore, unsuitable or inappropriate compression strategies can utilize more resources and energy than necessary. To that end, we propose a novel methodology for achieving on-the-fly intelligent decision making for energy-efficient data compression using machine learning. We have integrated a decision component into the Scientific Compression Library (SCIL) and show that, with appropriate training, our approach allows SCIL to select the most effective compression algorithms for a given data set without users having to provide additional information. This enables achieving compression ratios on par with manually selecting the optimal compression algorithm.

@inproceedings{IEEOSDCWDT20,
	author	 = {Michael Kuhn and Julius Plehn and Yevhen Alforov and Thomas Ludwig},
	title	 = {{Improving Energy Efficiency of Scientific Data Compression with Decision Trees}},
	year	 = {2020},
	month	 = {09},
	booktitle	 = {{ENERGY 2020: The Tenth International Conference on Smart Grids, Green Communications and IT Energy-aware Technologies}},
	publisher	 = {IARIA XPS Press},
	pages	 = {17--23},
	conference	 = {ENERGY 2020},
	location	 = {Lisbon, Portugal},
	isbn	 = {978-1-61208-788-7},
	issn	 = {2308-412X},
	abstract	 = {Scientific applications, simulations and large-scale experiments generate an ever increasing deluge of data. Due to the storage hardware not being able to keep pace with the amount of computational power, data reduction techniques have to be employed. Care has to be taken such that data reduction does not impact energy efficiency as it is an important cost factor for supercomputer systems and infrastructures. Data reduction techniques are highly data-specific and, therefore, unsuitable or inappropriate compression strategies can utilize more resources and energy than necessary. To that end, we propose a novel methodology for achieving on-the-fly intelligent decision making for energy-efficient data compression using machine learning. We have integrated a decision component into the Scientific Compression Library (SCIL) and show that, with appropriate training, our approach allows SCIL to select the most effective compression algorithms for a given data set without users having to provide additional information. This enables achieving compression ratios on par with manually selecting the optimal compression algorithm.},
	url	 = {https://www.thinkmind.org/index.php?view=article&articleid=energy_2020_1_40_30038},
}