Date of Publication :13th March 2017

Abstract: Resource providers on Cloud offer heterogeneous resources such as compute units and storage in Virtual Machine instances (VM). Cloud providers expect users to request for resources. In this process, overestimation of resources by cloud users lead to unused resources. The cumulative unutilized resources for each job of the user, amount to unnecessary expenditure for users and wasted resources for providers. Large scale data centers that provide reliable high performance computational and storage services for Cloud providers, face problems of increased energy consumption and CO2 emission as a consequence of huge resource wastage. Therefore, for environmental and financial reasons it is imperative to reduce unnecessary resource reservation. This can be done by using resource prediction technique that ensures resource allocation only as much as is necessary for the customer. Currently, there are no suitable prediction techniques for Cloud resource usage because of absence of pattern, trend and seasonality in users’ resource usage [2].The proposed prediction approach applies Enhanced Instance Based Learning and is tested using Google Cluster Trace Data [1] . The contribution of this work to existing body of work on cloud resource management by effective resource prediction approaches are discussed.

Reference :