Markov Chain Based Algorithm for Virtual Network Embedding in Optical Data Centers

• Published in: 2016 IEEE 18th International Conference on High Performance Computing and Communications; IEEE 14th International Conference on Smart City; IEEE 2nd International Conference on Data Science and Systems (HPCC/SmartCity/DSS) pp. 899 - 906
• Main Authors: Tram Truong-Huu, Gurusamy, Mohan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2016
• Subjects: Bandwidth; Cloud computing; Computational modeling; Data models; Markov processes; Markov Random Walk; optical data centers; Optical fiber networks; Optical switches; revenue maximization; virtual network embedding
• DOI: 10.1109/HPCC-SmartCity-DSS.2016.0129

Cloud data centers have become an attractive candidate for large-scale applications that require cloud resources in the form of a virtual network. Embedding virtual networks in data centers to allocate cloud resources to the applications running on the virtual machines is crucial since it affects resource efficiency and thus the revenue of cloud providers. Most of the existing embedding algorithms consider allocation of virtual machines and bandwidth as two sub-problems of the virtual network embedding problem and solve them separately. The performance drastically degrades when applying these algorithms to optical data centers due to the wavelength continuity constraint and high degree of the optical switch. In this paper, we propose a Markov Chain-based algorithm for virtual network embedding in optical data centers (MCA-VNE). We develop a Markov chain model to compute the ranking of a Top-of-the-Rack (ToR) based on not only its resource capacities but also that of its direct neighbors. The Markov model takes into account three resource parameters of a ToR: the computing capacity, the available bandwidth on the established lightpaths from itself to other ToRs and the possible bandwidth that can be provided in future based on the number of wavelengths currently available. We evaluate the performance of the proposed algorithm through comprehensive simulations. The results show that the proposed algorithm has better performance compared to baseline algorithms by reducing the rejection ratio by at least 10% and increasing the revenue for cloud providers by up to 13%.