Reward-based Markov Chain Analysis of Slicing Flows for Inter-Cloud Virtual Resources Allocation in 5G Cellular Network

Abstract

In 5G, diverse classes of traffic (RT/NRT) are differentiated as flows of network slices, e.g., uRLLC, eMBB, mMTC, etc., traversing 5G between the end users to global cloud data centers. Clearly, for minimizing cloud access delay and maximizing cloud net profit, several mechanisms are considered in 5G Cloud era: migrating virtual cloud resources, redirecting incoming request, steering traffic via mobile edge computing (MEC), overselling virtual resources, etc. However, the first three mechanisms increase the cloud cost, but the fourth one may increase cloud reward. With general business policy of overselling, the dynamic Service License Agreements (SLAs) are required for achieving the objective. To maximize system reward while satisfying the QoS of various SLAs in 5G, an efficient cloud resource allocation is required for cloud manager. The Reward-based adaptive global Cloud Resource Management, namely RCRM, is thus proposed for 5G. Moreover, RCRM consists of some contributions: Traffic Predictions, Adaptive Cloud resource Allocation and Maximum Net Profit. RCRM analyzes the service blocking and the required number of VMs for each request by the M/M/m/m Markov chain model. Generally, since cloud providers oversell diverse types of cloud resources, the objective of maximizing net profit can be achieved. Furthermore, the cost of deploying data centers at different areas in the world is completely varied, because in different areas (or countries) the equipment deploying cost, the power cost, the building cost, and the engineering cost are definitely different. For achieving 5G network slicing in different domains, diverse-type of cloud resources among global data centers are adaptively allocated in this work. From performance evaluations, the proposed RCRM approach outperforms the others in dropping probability, SLA violation, violation penalty and net profit. Moreover, the dropping probability is analyzed and the probabilities of analysis and simulation are very close. This result can justify the correctness of the proposed Markov chain model.