In recent years, the importance of using wireless network technology has increased. They have taken advantage of this technology, from smartphones and the Internet of Things to providing various cloud and edge computing services. With the growing use of this technology, it is necessary to provide things such as high reliability, low latency, quality of experience (QOS), and the possibility of connecting more devices simultaneously in the next generations. This generation tries to minimize the mentioned limitations by applying concepts and technologies such as network slicing, network function virtualization (NFV), software-defined networking (SDN), and multi-access edge computing (MEC). As a result, this technology is service-oriented, and any functionality can be defined as a service and provided to users. Considering the provision of services in microservices and the creation of chains of VNFs as software components of network functions, the placement of these services is one of the main challenges in this generation. Along with the optimal placement of the service chain in terms of resources, some other parameters, such as the delay in providing services, energy consumption, and the cost of energy consumption, should also be considered. Because 5G intends to provide services with low delay and cost to encourage users to use personalized services and provide more profit for service providers, previous researches show that most of the presented methods have not considered all these cases. For example, old methods have ignored the use of MEC, and in some other methods, only the service delay or energy consumption has been considered. In most research, the proposed method is only based on using optimization tools, such as CPLEX and MiniZinc, or heuristic and meta-heuristic methods have been used. These two categories of methods face a challenge regarding response time, and their compatibility with changing conditions is very low. Therefore, in this research, we have presented two methods, OGA and Hybrid-OGA. They are based on reinforcement learning methods and the encoder-decoder model. In the OGA method, we model the VNF-FGE problem as a binary linear programming problem by considering service delivery delay, service energy consumption, and energy consumption cost simultaneously. Also, there are two types of hosts with renewable energy and hosts with brown energy to choose from. Since some previous methods are not online, we have presented this method in an online environment. In learning methods, a model is usually trained first, and then that model is used, which is a time-consuming procedure. As a result, using the epsilon-greedy strategy, we have developed an intelligent agent based on reinforcement learning that improves the overall performance and the response time over time and the environment is updated at every step. Due to the greediness of this approach and the possibility of inappropriate blocking of some services, we have added a meta-heuristic approach based on a genetic algorithm to our method. This meta-heuristic approach uses the output of the same model to reduce the number of blocked services. Finally, we compared the proposed method with the First-Fit (FF) method and the results of the CPLEX tool. The eva‎luation results show that the OGA and Hybrid-OGA methods can compete well with the other two methods regarding response time, acceptance ratio, number of blocked services, and minimization of the objective optimization function.

زينب زالي

مسعودرضا هاشمي

محمدرضا حيدرپور , نغمه سادات مويديان