Contact at mumbai.academics@gmail.com or 8097636691/9323040215
Responsive Ads Here

Friday, 22 June 2018

Energy-Efficient Query Processing in Web Search Engines

Abstract— 

Web search engines are composed by thousands of query processing nodes, i.e., servers dedicated to processing user queries. Such many servers consume a significant amount of energy, mostly accountable to their CPUs, but they are necessary to ensure low latencies since users expect sub-second response times (e.g., 500 ms). However, users can hardly notice response times that are faster than their expectations. Hence, we propose the Predictive Energy Saving Online Scheduling Algorithm (PESOS) to select the most appropriate CPU frequency to process a query on a per-core basis. PESOS aims at process queries by their deadlines, and leverage high-level scheduling information to reduce the CPU energy consumption of a query processing node. PESOS bases its decision on query efficiency predictors, estimating the processing volume and processing time of a query. We experimentally evaluate PESOS upon the TREC ClueWeb09B collection and the MSN2006 query log. Results show that PESOS can reduce the CPU energy consumption of a query processing node up to ∼48% compared to a system running at maximum CPU core frequency. PESOS outperforms also the best state-of-the-art competitor with a ∼20% energy saving, while the competitor requires a fine parameter tuning and it may incur in uncontrollable latency violations.

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

INTRODUCTION

Web search engines continuously crawl and index an immense number of Web pages to return fresh and relevant results to the users’ queries. Users’ queries are processed by query processing nodes, i.e., physical servers dedicated to this task. Web search engines are typically composed of thousands of these nodes, hosted in large data centers which also include infrastructures for telecommunication, thermal cooling, fire suppression, power supply, etc. This complex infrastructure is necessary to have low tail latencies (e.g., 95th percentile) to guarantee that most users will receive results in sub-second times (e.g., 500 ms), in line with their expectations. At the same time, such many servers consume a significant amount of energy, hindering the profitability of the search engines and raising environmental concerns. In fact, data centers can consume tens of megawatts of electric power and the related expenditure can exceed the original investment cost for a data center. Because of their energy consumption, data centers are responsible for the 14% of the ICT sector carbon dioxide emissions, which are the main cause of global warming. For this reason, governments are promoting codes of conduct and best practices to reduce the environmental impact of data centers.

Since energy consumption has an important role on the profitability and environmental impact of Web search engines, improving their energy efficiency is an important aspect. Noticeably, users can hardly notice response times that are faster than their expectations. Therefore, to reduce energy consumption, Web search engines should answer queries no faster than user expectations. In this work, we focus on reducing the energy consumption of servers’ CPUs, which are the most energy consuming components in search systems.

 To this end, Dynamic Frequency and Voltage Scaling (DVFS) technologies can be exploited. DVFS technologies allow to vary the frequency and voltage of the CPU cores of a server, trading off performance (i.e., longer response times) for lower energy consumptions. Several power management policies leverage DVFS technologies to scale the frequency of CPU cores accordingly to their utilization. However, core utilization-based policies have no mean to impose a required tail latency on a query processing node. As a result, the query processing node can consume more energy than necessary in providing query results faster than required, with no benefit for the users.

In this work, we propose the Predictive Energy Saving Online Scheduling algorithm (PESOS), which considers the tail latency requirement of queries as an explicit parameter. Via the DVFS technology, PESOS selects the most appropriate CPU frequency to process a query on a per-core basis, so that the CPU energy consumption is reduced while respecting a required tail latency. The algorithm bases its decision on query efficiency predictors rather than core utilization. Query efficiency predictors are techniques to estimate the processing time of a query before its processing. They have been proposed to improve the performance of a search engine, for instance, to take a decision about query scheduling or query processing parallelization. However, to the best of our knowledge, query efficiency predictor have not been considered for reducing the energy consumption of query processing nodes

No comments:

Post a Comment