Performance Engineering in Cloud Computing

Abstract

Prior to the dawn of cloud computing, capacity planning was one of the most perplexing newlinetasks for software application developers. Cloud computing has driven a new age of newlinecomputing and service delivery model. Cloud computing enables cloud users access newlinevarious computing resources in the form of services from cloud service provider. newlineIn cloud computing, the developers need not pre-determine capacity planning with newlineregard to various computing resources like memory, processing and storage etc. newlineThe cloud users access resources from a cloud service provider according to changeable newlineworkload and bind to an agreement called Service Level Agreement (SLA). In order to newlinetackle all the users in cloud environment, the performance of the application needs to newlinebe measured in peak traffic. Poor performance of an application results in increased newlinecosts of software development and hardware and more importantly damaged customer newlinerelations. In particular, the performance assessment of cloud applications requires newlinespecial attention. In the current practice, constructing performance models of complex newlinesystems is expensive to develop and validate. Proven techniques or models are therefore newlineneeded for cloud applications for ease and accelerating the process of building and newlinesolving performance models. newlineThe thesis focuses on assessing the performance of cloud services and aims at designing newlinea performance prediction model, keeping in view of the research gaps of software newlineperformance engineering (SPE) with respect to cloud computing technologies, cloud newlinedata sizes, processing of cloud data, cloud service performance. A framework that newlinedescribes generic methodologies to implement the cloud service performance newlineprediction model is discussed in detail, in this thesis. An in-house experimental set up newlinein the laboratory is used to process various Twitter data sizes under Hadoop and Spark newlineplatforms.