MobiCoRE: Mobile Device Based Cloudlet Resource Enhancement for Optimal Task Response

Abstract

Cloudlets are small self maintained clouds, with hotspot like deployment, to enhance the computational capabilities of the mobile devices. The limited resources of cloudlets can become heavily loaded during peak utilization. Consequently, per user available computational capacity decreases and at times mobile devices find no execution time benefit for using the cloudlet. Researchers have proposed augmenting the cloudlet resources using mobile devices; however, the proposed approaches do not consider the offered service to load ratio while using mobile device resources. In this paper, we propose easy to implement Mobile Device based Cloudlet Resource Enhancement (MobiCoRE) while ensuring that: (i) mobile device always have time benefit for its tasks submitted to the cloudlet and (ii) cloudlet induced mobile device load is a fraction of its own service requirement from the cloudlet. We map MobiCoRE on M/M/c/K queue and model the system using birth death markov chain. Given the arrival rate of $\lambda$ , $c$ cpu cores in cloudlet, maximum tasks in the cloudlet to be $K$ and $P_0=f(\lambda, c,K,\mu)$ be probability of having no user in cloudlet, we derive the condition $\frac{1}{P_0}= \frac{\lambda ^K}{c^{K-c}c!\mu ^K}1000$ for optimal average service time $\frac{1}{\mu}$ of cloudlet such that the mobile applications have maximum benefit for using cloudlet services. We show that the optimal average service time is independent of the applications service requirement. Evaluation shows that MobiCoRE can accommodate upto 50 percent extra users when operating at optimal service time and sharing mobile resources for remaining task, compared to completing the entire user applications in cloudlet. Similarly, up to 47 percent time benefit can be achieved for mobile devices by sharing only 16 percent computational resources with the cloudlet.