Web QoE: Moving beyond Google's SpeedIndex

The World Wide Web is still among the most prominent Internetapplications. While the Web landscape has been in perpetual move-ment since the very beginning, these last few years have witnessedsome noteworthy proposals such as SPDY, HTTP/2 and QUIC whichcould disrupt the Web status quo and profoundly reshape the protocolsfamily at application layer. Technically solid means are clearly neededto assess whether these new protocols can keep their promises: Therisk is that these new protocols could otherwise fail to be adopted.While this investigation is already under way, both the industryand the research community are in our opinion expressing the rightquestion, to which they however answer using the wrong tools.Over the years, webpages have grown to quite complex entitiesincluding hundreds of objects of several types, sharded over manydomains. Yet, the current practice is to express Web Quality ofExperience (QoE) via the document completion time (onLoad) de-spite its known inaccuracy and poor correlation with the actual userexperience. At the same time, while better metrics do exist (e.g.,the SpeedIndex, proposed by Google in 2012), they are complex toevaluate and require a prohibitive amount of computing resources(i.e., record filmstrips of the visual rendering in the browser). Assuch, their use is limited to lab experiments, but have to date failedto catch up in larger scale.Fortunately, there is a way out of this impasse: as it often happens,once found, the solution is very simple and elegant.