MetaVRadar: Measuring Metaverse Virtual Reality Network Activity

TL;DR

The paper addresses how telecommunications networks can be made metaverse-ready by uncovering the complex, multi-domain traffic patterns of VR metaverse sessions and their impact on user experience. It introduces MetaVRadar, a three-stage, real-time detection and state-classification framework that leverages encryption-agnostic signatures from TLS handshakes and time-critical UDP flows, plus volumetric traffic attributes, using a stateful random-forest approach with past-state context. Empirically, the authors characterize traffic across six user activity states in four Oculus-based metaverses, validate high detection and classification accuracy in lab settings, and demonstrate practical deployment insights in a university campus network, including AS- and latency-aware implications for operators. The findings enable operators to optimize bandwidth, caching, and routing to support immersive metaverse experiences, while highlighting the need for periodic retraining as platforms evolve.

Abstract

The "metaverse", wherein users can enter virtual worlds to work, study, play, shop, socialize, and entertain, is fast becoming a reality, attracting billions of dollars in investment from companies such as Meta, Microsoft, and Clipo Labs. Further, virtual reality (VR) headsets from entities like Oculus, HTC, and Microsoft are rapidly maturing to provide fully immersive experiences to metaverse users. However, little is known about the network dynamics of metaverse VR applications in terms of service domains, flow counts, traffic rates and volumes, content location and latency, etc., which are needed to make telecommunications network infrastructure "metaverse ready". This paper is an empirical measurement study of metaverse VR network behavior aimed at helping telecommunications network operators better provision and manage the network to ensure good user experience. Using illustrative hour-long network traces of metaverse sessions on the Oculus VR headset, we first develop a categorization of user activity into distinct states ranging from login home to streetwalking and event attendance to asset trading, and undertake a detailed analysis of network traffic per state, identifying unique service domains, protocols, flow profiles, and volumetric patterns, thereby highlighting the vastly more complex nature of a metaverse session compared to streaming video or gaming. Armed with the network behavioral profiles, our second contribution develops a real-time method MetaVRadar to detect metaverse session and classify the user activity state leveraging formalized flow signatures and volumetric attributes. Our third contribution practically implements MetaVRadar, evaluates its accuracy in our lab environment, and demonstrates its usability in a large university network so operators can better monitor and plan resources to support requisite metaverse user experience.

Figures (10)

• Figure 1: Our lab measurement setup.
• Figure 2: Screenshots of example user activities and their states in one Metaverse (Multiverse) session.
• Figure 3: Complexity of metaverse user network activities in our representative walk-through: (a) service domains being accessed ranked by total packet count; and (b) distribution of flows across different user activity states, autonomous systems, and latency ranges.
• Figure 4: Flow span and volumetric patterns of our one example Multiverse session consisting of 17 user activities.
• Figure 5: Distribution of flows across autonomous systems (AS) and latency ranges for SUE, SPE, and AT states in our representative metaverse walk-through.