PROJECT TITLE :

Energy-efficient Query Processing in Web Search Engines - 2017

ABSTRACT:

Web search engines are composed by thousands of question processing nodes, i.e., servers dedicated to process user queries. Such several servers consume a important amount of energy, largely accountable to their CPUs, however they are necessary to ensure low latencies, since users expect sub-second response times (e.g., five hundred ms). However, users will hardly notice response times that are faster than their expectations. Hence, we have a tendency to propose the Predictive Energy Saving Online Scheduling Algorithm (\sfPESOS ) to pick out the most appropriate CPU frequency to method a question on a per-core basis. \sfPESOS aims at method queries by their deadlines, and leverage high-level scheduling info to scale back the CPU energy consumption of a question processing node. \sfPESOS bases its decision on query potency predictors, estimating the processing volume and processing time of a question. We experimentally evaluate \sfPESOS upon the TREC ClueWeb09B assortment and therefore the MSN2006 question log. Results show that \sfPESOS can reduce the CPU energy consumption of a question processing node up to \sim 48 p.c compared to a system running at maximum CPU core frequency. \sfPESOS outperforms conjointly the simplest state-of-the-art competitor with a \sim 20 p.c energy saving, whereas the competitor requires a fine parameter tuning and it could incurs in uncontrollable latency violations.