Abstract
In the face of rising global demand for video meetings, managing traffic across geographically distributed (geo-distributed) data centers presents a significant challenge due to the dynamic and limited nature of inter-DC network performance. Facing these issues, this paper introduces two novel techniques, VCRoute and WMJitter, to optimize the performance of geo-distributed video conferencing systems. VCRoute is a routing method designed for audio data packets of video conferences. It treats the routing problem as a Multi-Armed Bandit issue, and utilizes a tailored Thompson Sampling algorithm for resolution. Unlike traditional approaches, VCRoute considers transmitting latency and its variance simultaneously by using Thompson Sampling algorithm, which leads to effective end-to-end latency optimization. In conjunction with VCRoute, we present WMJitter, a watermark-based mechanism for managing network jitter, which can further reduce the end-to-end delay and keep an improved balance between latency and loss rate. Evaluations based on real geo-distributed network performance demonstrate the effectiveness and scalability of VCRoute and WMJitter, offering robust solutions for optimizing video conferencing systems in geo-distributed settings.
Introduction
In the post-pandemic era, video conferencing traffic has exploded, relying on geographically distributed data centers (DCs) to connect users globally. However, the "Internet backbone" between these DCs is notoriously fickle, characterized by dynamic and unpredictable performance. A critical insight of this work is that end-to-end latency is the sum of two distinct components: transmitting latency (time in flight) and reordering latency (time spent in a jitter buffer at the destination).
Existing solutions typically focus on minimizing transmission latency alone. However, a path that is fast on average might have high "jitter" (variance in delay). High jitter forces traditional jitter buffers to hold onto packets longer to reorder them, inadvertently increasing the total end-to-end delay. Furthermore, traditional buffers use "In-Order Processing" (IOP), where the entire stream stalls if a single "straggler" packet is late.
Methodology
To overcome these limitations, we propose a two-pronged approach: VCRoute for intelligent path selection and WMJitter for efficient jitter management.
VCRoute: Bayesian Path Selection
Instead of a simple "shortest path" algorithm, VCRoute treats routing as a Multi-Armed Bandit (MAB) problem. It utilizes a tailored Thompson Sampling algorithm to select paths. Unlike standard algorithms that only look at average speed, Thompson Sampling uses a Bayesian framework to model the probability distribution of a path's performance. This allows VCRoute to simultaneously consider both the expected latency and its stability (variance), effectively avoiding jittery paths before a packet even leaves the source.
WMJitter: Watermark-based Jitter Management
At the receiver end, we replace the standard jitter buffer with a Watermark-based Out-of-Order Processing (OOP) mechanism. Borrowing concepts from big data stream processing, "watermarks" are timestamps that signal the progress of the stream. If a packet arrives and its timestamp is older than the current watermark, it is discarded as too late. Crucially, WMJitter allows packets to be processed even if some are missing, provided the watermark indicates it is safe to move on, preventing the "head-of-line blocking" common in traditional buffers.
Evaluation and Results
The system was evaluated using real-world network traces from Tencent Cloud (simulating a managed backbone) and WonderProxy (simulating public Internet conditions across 250+ servers). These methods were integrated into a trace-driven simulator based on the industry-standard WebRTC framework.
Key results include:
- Latency Reduction: VCRoute and WMJitter together reduced end-to-end latency by up to 44% compared to baseline "Direct" routing and standard jitter buffers.
- Superiority over Speed-Centric Routing: Compared to Via (a state-of-the-art relay selection algorithm), VCRoute reduced average latency by 4% to 13% because it accounted for jitter, whereas Via only focused on transmission speed.
- Reliability in High-Jitter Environments: In chaotic network conditions, WMJitter reduced packet loss by 2-4% compared to traditional buffers while simultaneously lowering delay.
- Minimal Overhead: Despite being a centralized scheduler, VCRoute's overhead is negligible, accounting for less than 1% of the total end-to-end latency.
Conclusion
By shifting the focus from raw transmission speed to a holistic view of end-to-end latency that includes reordering time, VCRoute and WMJitter offer a robust solution for global video conferencing. The use of Bayesian path selection and out-of-order processing demonstrates that stability and timely delivery are paramount for real-time communication in geo-distributed settings.
Summary updated