On-Line Cost-Aware Workflow Allocation in Heterogeneous Computing Environments

Abstract

With the appearance of on-line big data stream computation, the explosive growth of mobile devices, the development of broadband cellular network, and widespread use of WiFi in recent years, the VM allocation problem has shifted gradually from batch processing to real-time processing. As the processing streaming workflow allocation becomes very large, it has become far more difficult. First, in this paper, we have modeled new network based on mobile cloud computing and mobile edge computing scheme for the real-time streaming workflow allocation problem. Our proposed network called Heterogeneous Node Network (HNN) consists of three types of computing node. HNN has a conventional data center (DC), a cloudlet (CL) located between edge server (ES) and DC, and ES consisting of mobile devices. In HNN, DC is the conventional placement destination of virtual machine (VM) and has high computing resource compared to other nodes; CL is a new computing resource, whose performance is lower than DC, but data transmission between CL and ES is faster than between DC and ES, and ES is a cluster of mobile devices with the lowest computing resource and its advantage is reducing the amount of data from raw data for crucial processes of streaming workflow. Second, we propose a heuristic streaming workflow allocation algorithm, which is flexible according to change of real-time availability for streaming workflow and HNN environment to achieve cost minimization. Our algorithm is the hybrid of a bin-packing algorithm and a shortest path algorithm based on the VM placement problem and the shortest path problem in graph network respectively. Finally, our developed algorithm has been compared with the result of linear programming (LP). In performance evaluation, the experimental results show our approach leads to a solution close to an optimal solution generated by LP and its execution time is reduced.