Machine Learning as a Means to Adapt Requirement Changes for SDN Deployment Process in SDN Migration

摘要

The deployment of SDN in legacy network has gained popularity across network operators as next generation network architecture. However, full deployment of SDN faces challenges in economical, organizational, and technical aspects. Hence, the deployment of SDN should be incremental over months or even years, and limited numbers of the nodes are upgraded to SDN-enabled one in each period. This forms a hybrid SDN (H-SDN) network which legacy and SDN nodes co-exist in the same network. Importantly, which and when a node should be replaced to SDN node are the common question which impacts the performance of a hybrid SDN network. Efforts to date primarily focus on determining sequence for migration which maximize the performance of traffic engineering (TE) in H-SDN network. However, most works do not take into consideration of the changes that may happen over the periods of SDN deployment. The possibility of these changes requires adaptation techniques to ensure effective migration sequence to cater present and future needs. In this article, we aim to identify the gap and propose the opportunity in which techniques originated from machine learning (ML) may play an important role in solving problem of incremental SDN deployment by alleviating the issues the occur during SDN migration as well as to improve the H-SDN deployment.