Measuring Human Satisfaction in Data Networks

There is currently no widely accepted method for converting packet-based measurements of Quality-of-Service (QoS) into a score reflecting a user's opinion of network performance. Such a measure is desirable for comparing networks, making design trade-offs, or enforcing service-level agreements (SLAs), and is likely to be a popular area of future study. It is of particular importance in wireless networks where network resources are especially limited. To build a user-centric measure of performance, it is necessary to have a robust method for measuring user opinions of web surfing which can be directly related to packet-based measures. In this paper we present a methodology that is similar in function to the Mean Opinion Score (MOS) in the voice domain, and is therefore dubbed the dataMOS. In our method, users actively engage in a series of realistic internet tasks over emulated wireless connections varying in bandwidth and propagation delay. For the majority of the paper we focus on web browsing on a laptop. For each task, users rate the experience in terms of overall quality, perceived browsing speed, and website design. Results indicate several interesting phenomena: * Overall satisfaction is heavily influenced by website design and the task being performed and is not only affected by perceived network speed. * User ratings of connection quality increase with bandwidth until they reach a maximum at roughly 256--400kbps. We dub this point the saturation bandwidth. To capture this effect we model user opinions at given bandwidths with a ``slope-saturation'' model that conveniently captures both the quality/bandwidth trade- off for a given task, as well as the saturation bandwidth. * Propagation delay does not have a significant effect on user satisfaction since, for all the tasks and bandwidths considered, queueing delay dominates any reasonable propagation delay.