A migration method for service function chain based on failure prediction

With the deep application of network technologies in different industries, the demand for network services is becoming more and more diversified. Network operation and maintenance are facing severe challenges, which can be solved by network function virtualization (NFV). NFV technology provides services for users through deploying service function chain (SFC) on servers and substrate links. However, once a server fails, the services it hosts will be affected or even interrupted. Therefore, it is very important to predict failures and migrate SFCs in advance according to predicted results. In this paper, we propose a failure prediction method based on the improved long short-term memory neural network (PMILSTM), which employs LSTM to predict failures. To further improve prediction accuracy, the simulated annealing particle swarm optimization algorithm is adopted to optimize the number of neurons in each long short-term memory layer and the time window length. A server may host multiple SFCs. When a server fails, in order to reduce the impact on users, it is necessary to simultaneously migrate all the SFCs hosted by the server. We propose an improved sparrow search algorithm (ISSA) and a service function chain migration method based on the ISSA (MMISSA). The ISSA introduces tent chaos, opposition-based learning, dynamic weight factor, and mutation operation into SSA to achieve the better global optimization ability. The MMISSA method adopts the ISSA to migrate SFCs so that it can simultaneously search for migration servers for all the virtual network functions deployed on a soon-to-fail server. The better global optimization ability of the ISSA enables better migration results. Therefore, the migration success ratio is improved. Moreover, the fitness function simultaneously considers the average migration cost and migration time. As a result, the MMISSA method effectively reduces the migration cost and migration time.