An efficient elastic resource management in cloud computing environment using Prediction approaches

Abstract

In cloud computing, resource management manages the cloud resources, and the newlinemain objective is to allocate the computing resources to the applications in time with newlineminimal cost. A workload prediction module occupies a key position in the intelligence newlineof resource management systems. However, due to the high fluctuations in incoming newlineworkloads, workload prediction has become a very complex task. Inaccurate workload newlineprediction leads to unwanted scaling operations for resources and leads to SLA violations, newlinehigh power consumption, and cost. In existing works, conventional prediction newlinealgorithms were used to estimate the workload of the applications. However, these algorithms newlinecan give a poor accuracy rate in predicting irregular workload patterns. This newlinedissertation presents the study of predictive approaches to achieve high performance in newlineresource allocations. In this context, the first proposed cost aware cloud workload management newlineframework handles the non-linear workload patterns efficiently by using an improved newlineand optimized LSTM network. The principal aim of this proposed framework newlineis to allocate suitable and accurate VM resources in advance to execute future workloads newlineso that service providers can avoid the unnecessary cost of undesired resource newlineallocations. The second proposed an efficient cloud resources utilization framework newlinethat utilizes the CPU and memory resources efficiently by using a dynamic threshold newlinebased auto scaling method and multi-resource usage prediction method. The major newlineaim of this proposed framework is to utilize cloud resources effectively while avoiding newlineunder-provision issues. The third proposed a resource, power, and SLA aware VM consolidation newlinetechnique to place the VMs in suitable host machines by using the proactive newlinehost utilization detection method, multi-objective VM selection, and multi-objective newlineVM placement approach. The major aim of this proposed framework is to reduce data newlinecenter power consumption and avoid SLA violations. The three proposed frameworks newlinewere tested with benchmark workl