Publication – Deep Learning for Proactive Network Application Management

Outline the paper

We are witnessing an emergence of a new era of applications delivered via a paradigm of flexible and softwarized communication networks. This has opened the market to a wider movement towards virtualized applications and services in key verticals such as automated vehicles, smart grid, virtual reality (VR), Internet of Things (IoT), industry 4.0, telecommunications, and more.

With an increasing emergence of verticals driven by the vision of low latency and high reliability, there is a wide gap to efficiently bridge the Quality of Service (QoS) constraints for the end-user experience. Most latency-critical services are over-provisioned on all fronts to offer reliability, which is inefficient in the long run. In this work, we present a Residual Long Short-Term Memory (LSTM) based multi-label classification framework for proactive SLA management in a latency-critical Network Function Virtualization (NFV) application use case.

We compose a multivariate time-series forecasting model with multiple time-step predictions in a multi-output scenario, and associate a multi-label classifier for a granular prediction of individual Service Level Objective (SLO) violations for each step in the forecast horizon. The Residual LSTM approach achieves an improvement of 31.1% over the baseline on the forecast classification accuracy, and a 2.65% improvement on the interpolated average precision over the standard LSTM methodology.

Who will it help?

Approaches like this are targeted towards efficiently improving the end-user experience while delivering latency-critical applications and services in verticals such as automated vehicles, smart grid, virtual reality (VR), Internet of Things (IoT), industry 4.0, and telecommunications. Most latency-critical services are over-provisioned on all fronts to offer reliability, which is inefficient towards scalability in the long run. Over time, we need to be mindful of the significant carbon footprint of networks and algorithms too, so there is more to it than just the end user experience here.

With the use of advanced machine learning, we can control how the Cloud reacts to such a time-sensitive demand, mitigate any disruptions before they impact the end-user, and ensure that systems and services are precisely proactive over time. We propose a deep learning based solution towards adequately balancing the efficiency and reliability in such a setup, towards supporting latency-critical applications that demand high availability values.

What is the future of this research?

We demonstrate the suitability of a Residual LSTM model over other Deep Neural Network and Recurrent Neural Network based methodologies in such a scenario that involves fine-grained rapid forecasting, and reason that the high level of granularity in predicting Service Level Objectives as multi-label outputs would help ensure a balance in precise provisioning while maintaining reliability in latency-critical NFV applications.

The methodology is transferable to other verticals within the 5G and 6G high-availability network slice, and in our future work we plan to validate this on a different use-case. We also plan to extend the deployment to a bigger test-bed setup, aiming to incorporate the external network features by setting up the Software Defined Networking block, and distributed application scenarios.

Publication Title: A Residual LSTM based Multi-Label Classification Framework for Proactive SLA Management in a Latency Critical NFV Application Use-Case
Authors: Nikita Jalodia, Dr Mohit Taneja, Dr Alan Davy, Dr Behnam Dezfouli
Conference Dates: 08-11 January 2022
Publication Date: 10 February 2022
Name of Conference: 2022 IEEE 19th Annual Consumer Communications & Networking Conference (CCNC)
Link to publication: https://doi.org/10.1109/CCNC49033.2022.9700502
Link to video presentation: https://whova.com/portal/ccnc1_202201/videos/2YzM0kzMzczN/