Distributed LSTM-based cloud resource allocation in Network Function Virtualization Architectures

Abstract

The high reconfiguration time of cloud resources in Network Function Virtualization environments has led to the proposal of prediction-based resource allocation algorithms, with extensive use of artificial intelligence techniques. The prediction of processing capacities performed jointly and centrally has proved to be very complex due to the high communication overhead required. For this reason, we propose a distributed prediction technique in which Long Short-Term Memory neural networks exchange only a few weights in order to drastically reduce the communication overhead compared to the centralized case. We propose and investigate three different distributed solutions and show how they allow for low prediction errors.