RELATED WORK In this section, we discuss various aspects of mobile cloud computing in general. Then we focus on game-specific literature on this field. A. Mobile Cloud Computing Mobile Cloud Computing [1] research aims to bring rich computing applications to energy constrained hand-held devices, such as mobile phones and tablets. While most of the hand-held devices are quite capable nowadays, the battery technology has not improved at the pace of processor and memory technology. According to some estimations [2] doubling the clock speed of the processor consumes eight times more energy, which makes running computationally intensiveapplications,suchasimageorvoicerecognition,very challenging at the mobile end. As the processor is a large constraint for mobile devices, approaches to offload computation from the devices to servers (with fixed power supply) have emerged. A classic and a very coarse-grained approach to offload computation is to use the mobile as a “thin client”, i.e., use the mobile client as a dumb display and control device, and handle all of the remaining computation at the server side. More recent approaches allow offloading on-demand and using more fine-grained granularity (e.g. process, method or class) [2]. Such approaches are referred as mobile computation offloading, cyber foraging or surrogate computing in the literature. All of the mobile computation offloading systems either aim to save energy of the mobile device or make it possible to accomplish tasks that are not normally possible solely using
2016 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS): GI 2016: 9th IEEE Global Internet Symposium
978-1-4673-9955-5/16/$31.00 ©2016 IEEE
the mobile device. The offloaded tasks should be carefully chosen because moving the computation unit and related data over wireless networks consumes also energy. As a rule of thumb, the offloaded task should be computationally intensive and the amount of data should be minimized in order for the offload operation to be justifiable [2]. If the data is already available at the server side, this naturally saves energy (e.g. for searching a face from a set of preloaded pictures). In addition, the underlying network type (802.11x vs. 3G vs. 4G) also impacts the energy consumption [3]. The cloudlet architecture from Carnagie-Mellon university [4] represents a well-know approach for mobile cloud computation. The granularity in cloudlets is a virtual machine (VM); a mobile device can start a service-specific VM in the cloud. To minimize data transfer, the cloud already contains a base image (e.g. basic Ubuntu Linux installation) and the device just transfers a delta to the image (e.g. Apache web server and its configuration). The “cloud” in cloudlet architecture is two-tiered: as the first tier, a miniature datacenter (i.e. a “cloudlet”) may be deployed near the device to reduce offloading latency and, as the second tier, large-scale datacenter located further away may be used in the absence of such. It should be noted that the cloudlet implementation architecture is built on hypervisor-based virtualization, and this leaves room for improvement in form of light-weight virtualization (e.g. Docker-based containers) that can improve the resource utilization at the server side [5]