Scalability analysis and performance evaluation of WebRTC systems using the Infrastructure as Code paradigm and containerized traffic generators

The dynamic development of remote education and distributed work has driven the demand for reliable real-time communication platforms.Ensuring the stability of such systems under massive load is a standard engineering challenge, necessitating robust methodologies for scalability assessment and proactive cloud infrastructure monitoring.Despite substantial progress in WebRTC benchmarking, practical gaps remain in the context of high-concurrency educational deployments.To achieve rigorous experimental repeatability and eliminate the configuration drift common in traditional load testing, the research adopted an advanced evaluation framework based on declarative infrastructure provisioning (Infrastructure as Code).The target environment for this evaluation was the BigBlueButton (BBB) system (version 3.0.19),deployed on a high-capacity server (e.g., 64 vCPU, 315 GiB RAM) as part of a major educational project in Poland.The traffic generation environment consisted of a distributed cluster of Docker containers, ensuring full isolation and repeatability of test scenarios.End-user simulation (utilizing headless Chromium clients and standard video resolution profiles) was implemented using scripts based on the Playwright library, allowing for the emulation of realistic interactions with the WebRTC interface.Iterative load tests at a scale of up to 421 users demonstrated the high efficiency of the server cluster, which provided empirical evidence that backend compute resources do not constitute the exclusive limiting factor in massive-scale deployments.The analysis revealed that the critical limitation of the system is the performance of the client-moderator's browser, which loses functionality beyond the threshold of 250 active video streams despite significant server resource redundancy.Quantitative metrics confirmed this imbalance: at peak load, server CPU utilization hovered between 12.1% and 16.2%, and RAM usage reached only 10 GiB out of 315 GiB, whereas external network traffic peaked at 374 Mb/s with a noticeable rise in TCP retransmissions and packet drops.The results confirm that the analyzed infrastructure is capable of stably supporting user groups that significantly exceed the initial design assumptions.The research suggests that for sessions exceeding 250 participants, engineering efforts should focus on client-side application optimization and local workstation resource management, rather than merely scaling server resources.