AVIRA: Enhanced Multipath for Content-aware Adaptive Virtual Reality

Abstract

This paper presents Adaptive VR (AVIRA), a scheme that implements a Virtual Reality (VR) content-aware prioritisation transport to extend Multipath TCP (MPTCP) functionalities and improve its performance. To do so, AVIRA monitors the subflows operation and forecasts subflows' performance by applying an Machine Learning (ML) approach to evaluate a set of features -such as latency and throughputfor every subflow available. This ML approach forecasts the performance of these features through linear regression and applies a linear classifier by using a weighted sum on the forecast results. When the traffic of a specific VR component is detected, AVIRA performs its prioritisation scheme by redirecting packets to the subflow with the best set of forecasted features. AVIRA outperforms the algorithms used for comparison and shows that the use of an ML approach in a "low-level" application is viable, especially in situations where the network features under scrutiny are subject to higher variations. In these scenarios, the AVIRA scheme can be outstandingly efficient.

Learning to Harness Bandwidth With Multipath Congestion Control and Scheduling

“…More recently, advances in machine learning have been applied to the design of MPTCP [3], [8], [15], [29]- [32], [36]. Mai et al [29] applied deterministic policy gradient for learning the optimal congestion control strategies.…”

“…Liao et al [30] proposed a DRL-based multipath scheduler aiming at decreasing the out-of-order queue sizing under heterogeneous paths. Silva et al [32] implemented an adaptive virtual reality with content-aware prioritisation to improve MPTCP's performance.…”

“…Both of these designs [3], [8] focus on congestion control and do not address packet scheduling. Indeed, all of the aforementioned MPTCP designs [3], [8], [15], [29]- [32], [36] considered either congestion control or packet scheduling separately (instead of jointly). It is known that the performance of MPTCP is significantly impacted by packet scheduling which decides what packet to send on which path among the available paths that have open window.…”

“…The advantage of the MPTCP controller in [8], however, is shown only when it is applied to a group of sources, rather than the usual and versatile case of a controller per source. In addition, the stability of the proposed algorithms [3], [8], [29]- [32] are not addressed, whereas stability is a key desirable goal in the original MPTCP design philosophy [10]. Authors in [15], [36] introduced Deep Q Network (DQN) framework to enhance the MPTCP congestion control [36] and packet scheduling [15] performance in asymmetric paths.…”

Latency and Information Freshness in Multipath Communications for Virtual Reality

“…Furthermore, scheduling packets over the available paths is another problem specific to multipath transmission [22], which can compound with the congestion issues and interact with congestion control in non-obvious ways [23]. As it is the most common protocol, however, MPTCP is used in a number of VR solutions [24], which devise application-level schemes to work around its issues.…”