NSGA-II Based Multi-Objective Disaster Recovery Scheduling for Virtual Cloud Platforms
This study proposes a multi-objective optimization (MOO) method based on the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to improve the virtual cloud platforms' disaster recovery scheduling efficiency. First, an MOO model is constructed. The model defines the resource parameters of physical nodes and virtual machines. Meanwhile, it designs a three-objective function to "minimize disaster recovery response time, maximize resource utilization, and minimize costs". Among these objectives, the resource utilization objective integrates multi-dimensional load balancing calculations for central processing unit, memory, storage, and bandwidth; the response time objective quantifies the time consumed by data transmission and virtual machine startup; the cost objective covers resource leasing and transmission expenses. At the same time, constraints related to resource capacity, virtual machine uniqueness, compatibility, and data consistency are incorporated into the model. For algorithm implementation, binary encoding directly represents the virtual machine-to-physical node allocation relationships xij. The design incorporates simulated binary crossover with a probability of 0.9 and polynomial mutation operators with a probability of 0.1, both adapted for virtual cloud environments. A selection mechanism of "non-dominated sorting + elite retention" is adopted. The solution process is optimized by combining the dynamic characteristics of disaster recovery scenarios (real-time update of resource status and dynamic adjustment of disaster levels). Threshold verification is used for resource capacity constraints; a hierarchical feedback method is applied to adjust the allocation strategy for data consistency constraints (which rely on the virtual machine delay difference |Ta-Tb| ≤ δ), ensuring the proportion of feasible solutions. The experiment simulates a large-scale cloud environment based on Google Cluster Data, setting three scenarios: small-scale node failure, large-scale regional disaster, and mixed failure. The proposed method is compared with the Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) and NSGA-III. The results show that NSGA-II achieves the optimal load balance degree. In the small-scale failure scenario, the load balance degree is 21.1% and 19.0% lower than that of MOEA/D and NSGA-III, respectively. In the large-scale disaster scenario, it is 35.7% and 25.0% lower. In the large-scale scenario, the response time of NSGA-II is 15.2%-28.3% shorter than that of the benchmark algorithms; its cost is 22.8% lower than that of MOEA/D (with significant optimization in resource leasing cost). Compared with previous studies, the innovations of this study are as follows. At the modeling level, it breaks through the single-dimensional load optimization of traditional post-disaster scheduling and adapts to the virtualization characteristics of cloud platforms. At the algorithm level, it solves the problem of insufficient dynamic adaptation of traditional NSGA-II in virtual cloud disaster recovery through scenario-based encoding and constraint processing. At the practical level, it fills the method gap between disaster recovery scheduling in virtual cloud scenarios and that in traditional physical scenarios. This study enriches the application of MOEA in cloud resource management and provides theoretical and technical support for improving the disaster recovery capability of cloud platforms.
Read more