An Efficient Cache Eviction Strategy based on Learning and Belady Algorithm

With the rapid development of 5G networks and the explosive growth of mobile traffic demand, the transmission of massive data has exacerbated the burden on the backbone network, leading to an increase in network latency. Edge caching can bring caching functions to the edge of the network, providing users with the desired content in proximity, thereby reducing the pressure on the backbone network and lowering network latency. However, the limitations of edge server performance and cache capacity necessitate an efficient caching strategy to improve cache hit ratio. In this paper, we propose a new caching strategy (ECELB) that utilizes the concept of the offline optimal Belady algorithm to evict the object with the farthest next request time. The scheme is based on gradient boosting machine (GBM) and long short-term memory network (LSTM) to achieve filtering of object requests and prediction of the next arrival time of cached objects. The simulation results show that ECELB algorithm not only significantly reduces the computation cost of the model but also ensures its cache hit ratio.