Max Downey, General Manager Network Traffic Modelling & Insights at NBN, shared an in-depth empirical study on how network and service congestion impact perceived application Quality of Experience (QoE). His presentation explored attempts to quantify customer experience, focusing on whether faster speed plans translate to meaningful benefits.
Downey examined whether upgrading to higher speeds improves user experience, highlighting that faster speeds allow users to download large files, such as game patches, more quickly, reducing overall network strains by minimizing the duration of high-bandwidth activity.
Downey distinguished between two types of network congestion. Inter-service congestion occurs when multiple users share bandwidth on a network segment, while intra-service congestion happens within a single user’s dedicated connection. His study focused on the latter, particularly the limitations of varying speeds when handling multiple simultaneous video streams.
He identified two types of network traffic: firstly, elastic traffic – such as file downloads – expands to fill available bandwidth. Adding multiple downloads splits capacity, slowing each one proportionally. 
Inelastic traffic on the other hand, like video streaming or voice calls, has fixed bandwidth needs. If bandwidth is exceeded, additional streams are either blocked or experience degradation.
Experiments into inelastic traffic performance, showed that a 50 Mbps connection could comfortably support two 4K Netflix streams, but performance degraded with three. At 100 Mbps, networks could handle up to eight 4K Netflix streams without major issues.
Downey pointed to Netflix’s adaptive streaming behavior, which adjusts video resolution based on network conditions. He also discussed four causes of network latency: propagation delay, transmission delay, processing delay, and queueing delay.
To quantify latency’s impact on real-time applications, Downey explained how NBN used AI models trained on a competitive first-person shooter game (Quake 3/Open Arena) to reveal a clear decline in performance with increased latency but not significant impact from packet loss. The next phase of research will involve human players, as AI models lack adaptive flexibility in response to latency challenges.
Downey’s research provides key insights into how speed tiers affect customer experience and underscores the importance of optimizing network performance for different types of internet usage.