A Cost Cap-based Dynamic Resource Allocation Approach in Heterogeneously Multi-Experimental Data Computing Environments

Abstract

In scientific research environments, improving resource sharing and utilization is important for scientific data processing. Computing environments in scientific communities are in general dedicated to specific and mission-oriented experiments. Such approaches will lead to under resource utilization. Cloud computing is being gradually and actively adapted by scientific research communities because of its flexibility and promptness by on-demand approach. In order to improve utilization, there are two main factors seriously considered, which are cost and processing time. Cost and processing time are mutually conflict such that one can negatively impact the other. Therefore, it is important to find a balancing point between these two factors. In this paper, we propose a simple virtual machine allocation approach which can reduce computing complexity of load balancing problem by simply capping the cost. In our experiment, results show that our approach can be used to improve resource utilization as well as cost efficiency in multiple scientific data processing environments.