Cloud Load Prediction for Multiuser Mixed Distribution Task Length and Task Arrival Scenario

Abstract

In the mathematics queuing theory is one of the fields which can be used for finding solutions of a wide range of practical problems. One of the recently growing application of this field is in computer network traffic management. In this paper we applied the queuing theory for the prediction of job request queue in a cloud network where the job requests are arriving through a single queue and are combination of jobs generated by a number of users with each having the different job request patterns (distribution function). The growing adaption of cloud systems, resulting in excessive loading of cloud servers. Since the cloud is designed to provide different resources as service over the internet. The cloud user requests required resources from the cloud and the cloud serves these requests by forming virtual machines and allotting it some the resources from cloud resource pool. The resources allotted to VMs depends upon the requirements of the task and guaranteed QoS of the cloud. Once the task is completed these VMs can either be dissolved to utilize its resources for another VM or can be suspended for later utilization and saving power. The creation of a VM may take several seconds while activation of a suspended VM is also required some time, which may ultimately cause sluggish response and latency in cloud response. Such situations can be avoided by predicting the upcoming task requirements. The knowledge of upcoming tasks can be used to decide whether VMs needs to keep alive, suspended or dissolved to efficiently utilize the resources, saving power and serve the requests with minimum latency. The prediction of task requirements is a difficult job especially when the task requests are generated by different users and each user may follow different distribution function. This paper presents the HMM-based prediction model for such conditions and evaluates its performance.