Reduction of wasted energy in a volunteer computing system through Reinforcement Learning

Abstract

(2014) 'Reduction of wasted energy in a volunteer computing system through Reinforcement Learning.', Sustainable computing : informatics and systems., 4 (4). pp. 262-275. Further information on publisher's website:http://dx.doi.org/10.1016/j.suscom.2014.08.014Publisher's copyright statement: NOTICE: this is the author's version of a work that was accepted for publication in Sustainable Computing: Informatics and Systems. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reected in this document. Changes may have been made to this work since it was submitted for publication. A denitive version was subsequently published in Sustainable Computing: Informatics and Systems, 4, December 2014, 10.1016/j.suscom.2014.08.014. Additional information:Use policyThe full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-prot purposes provided that:• a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders.Please consult the full DRO policy for further details. AbstractVolunteer computing systems provide an easy mechanism for users who wish to perform large amounts of High Throughput Computing work. However, if the Volunteer Computing system is deployed over a shared set of computers where interactive users can seize back control of the computers this can lead to wasted computational effort and hence wasted energy. Determining on which resource to deploy a particular piece of work, or even to choose not to deploy the work at the current time, is a difficult problem to solve, depending both on the expected free time available on the computers within the Volunteer computing system and the expected runtime of the work -both of which are difficult to determine a-priori. We develop here a Reinforcement Learning approach tosolving this problem and demonstrate that it can provide a reduction in energy consumption between 30% and 53% depending on whether we can tolerate an increase in the overheads incurred.

Trace-Driven Simulation for Energy Consumption in High Throughput Computing Systems

“…These patterns are, however, quite complicated hence we evaluate the use of Reinforcement Learning in order to determine when and where jobs should be allocated within the institution in order to minimise the number of evictions and minimise the energy consumption. Through this work we have shown that it is possible to save up to 53% of the wasted energy within an institution [26].…”

“…These may include random allocation, lowest energy consumption, least chance of eviction, or fastest resource. Further discussion of these policies can be found in [25], [26].…”

HTC‐Sim: a trace‐driven simulation framework for energy consumption in high‐throughput computing systems

“…The HTC system must select the most appropriate candidates from a selection of avialable resources, in order to optimise the required performance, energy and QoS metrics. Many approaches exist including random allocation, reduced energy consumption, reduced likelihood of eviction, or fastest execution time [56,57,58] -discussed further in Section 5.5. The Pluggable Policy interface provides a simple interface which a developer can use to implement their own scheduler.…”

“…When considering the HTC workload, we may observe that the offered workload to our system in 2010 results in very low system utilisation (12%). BECK BRAE BRIG CHART DENE ELDON FAYOL FELL GATE GILL GLOBE HULL ISAAC LAKE LAWN LINN LOCH MOSS NAIAD NEREID NIDD ORACLE PARK PETH POND POOL SIDE TARN TEES TREE TURF TYNE WEAR WOOD Figure 8 shows the probability that a job of length x hours will complete given that it is submitted during hour y of the day [57]. Probabilities are obtained through simulation based on our Newcastle University trace logs for interactive users, and knowledge of computer reboots.…”

Using Machine Learning in Trace-driven Energy-Aware Simulations of High-Throughput Computing Systems

“…Request permissions from permissions@acm.org. computers least susceptible to preemption by their primary users [20], or removing tasks which are failing to complete (due to misconfiguration or broken code) [22].…”