Energy Efficient Resource Usage in Data Centers: GreenDoop

Energy Efficient Resource Usage in Data Centers: GreenDoop D. Fernández, A. Fernández-Montes, Juan A. Ortega Departamento de Lenguajes y Sistemas Informáticos Universidad de Sevilla Spain [email protected], [email protected], [email protected] Abstract The spread of cloud computing has increased recent data center size and usage. Because of this, the op timization of data centers is becoming the focus of most important companies in this sector. In order to improve data cenler energy efficiency, we propose a software infrastructure for minimizing power us age of data centers based on policies for switch ing on/off servers, load balancing and data organi zation; keeping the mínimum number of machines on, while affecting as little as possible performance and reliability. 1 Introduction The growing interest in energy efficiency has led big com panies to enhance its efforts in reducing power consump tion of data center infrastructures due to the increase in data centers energy consumption in the past ten years, which is 1.5% of worldwide energy consumption [Koomey, 2008; 2011]. Thanks to these efforts and researchers results, cur rent energy consumption trends have slowed down in the past five years. Many of these efforts have becn directed to ar eas unrelated to software, such as cooling [EI-Sayed, 2012), hardware scaling [Fan, 2007) or power distribution [Fema), 2005). This work, in contrast, focuses on software systems that support data centers, such as parallel computing systems or distributed file systems in order to develop a software prod uct that rninimizes the energy consumption. 2 Summary of Proposals In order to create a real software system we have to choose an existing distributed computing framework to work with. That system is Apache Hadoop, and our goal is to modify it by cre ating a software solution to coordinate and minimize energy consumtion at data center leve) by applying the designed en ergy efficiency policies to its two major modules: HDFS and YARN. 2.1 Modifications to HDFS HDFS is responsible for serving data. In order to minimize the number of machines running needed to maintain redun dancy benefits and data reliability we will apply different Actas de las XV Jornadas de ARCA. ]ARCA 2013 Sistemas Cualitativos y sus Aplicaciones en Diagnosis, Robótica e Inteligencia Ambiental J. Sanz, Ll. Museros, J. A. Ortega, A. Fernández-Montes (eds.) ISBN: 978-84-616-7622-4 © 2013 modifications: Restructuration of the Data Center to cate gorize block replicas, Design a new Balancing Policy to in crease server use and Create a Buffer and Caching System to handle versioning issues, prevent inconsistencies, hard ware errors and bad response times. 2.2 Modifications to YARN The purpose of YARN is running MapReduce applications and resource management [Tomasic, 2012). According to queuing theory, the arrival of events to a set of resources is described by exponential distribution and the arrival of con secutive events to a set of resources is described by a Gamma distribution. Our proposal is to start from these distributions to fit the situation: to minimize energy consumption. 3 Expected Results GreenDoop has the objective creating a highly configurable and useful enterprise too] that can notably reduce current Hadoop based data centers energy consumption. References [EI-Sayed, 2012) Nosayba El-Sayed. Temperature manage ment in data centers: Why sorne (might) like it hot. Tech nical report, Department of Computer Science, University of Toronto, 2012. [Fan, 2007) Xiaobo Fan. Power provisioning for a warehouse-sized computer. Technical report, Google Inc., 2007. [Femal, 2005) M.E. Femal. Boosting data center perfor mance through non-uniforrn power allocation. In Auto nomic Computing, 2005. ICAC 2005. Proceedings. Second International Conference on. Dept. ofComput. Sci., North Carolina State Univ., 2005. [Koomey, 2008) Jonathan Koomey. Worldwide electricity used in data centers. Environmental Research Letters, 2008. [Koomey, 2011) Jonathan G. Koomey. Growth in data center electricity use 2005 to 201 O. Technical report, Stanford University, 2011. [Tomasic, 2012) Ivan Tomasic. Bigdata and mapreduce with hadoop. In Proceedings of the I st International Confer ence on Cloud Assisted ServiceS, 2012. 95