Web search engines comprise numerous query processing nodes, which are servers explicitly engineered to manage user requests. A multitude of servers necessitates a significant amount of energy, largely due to their central processing units (CPUs). These servers are essential for ensuring minimal delays, as consumers expect response speeds under one second, specifically below 500 milliseconds. However, individuals find it difficult to recognize response times that surpass their established expectations. We present the Predictive Energy Saving Online Scheduling Algorithm (PESOS) to ascertain the best CPU frequency for processing individual queries at the per-core level. The main goal of PESOS is to effectively manage requests within designated deadlines, employing advanced scheduling data to reduce CPU energy consumption in a query processing node. PESOS assesses query efficiency by analyzing processing volume and the time necessary to execute a query. The experimental assessment of PESOS is performed on the TREC ClueWeb09B dataset and the MSN2006 query log. The results demonstrate that the application of PESOS results in a substantial decrease in CPU energy consumption for a query processing node, achieving around 48% compared to a system functioning at the maximum frequency of the CPU cores. PESOS outperforms the main state-of-the-art competition, attaining an estimated energy savings of 20%. Conversely, the competitor requires careful parameter calibration and may lead to erratic latency violations.

Energy consumption, CPU Dynamic Voltage and Frequency Scaling, Web search engines.