You Get What You Measure: Internet Performance Measurement as a Policy Tool

My paper on the disconnect between improvements in US broadband speed and the stagnation of web was published by the American Enterprise Institute today. Here’s the executive summary:

The Federal Communications Commission (FCC) justified its 2015 reclassification of internet service providers (ISPs) as common carriers with a novel “virtuous circle” hypothesis. The agency correctly observed that broadband networks enable innovation in platform services such as Facebook, Amazon, and Netflix. It then speculated that platforms cause improvements in network performance through a feedback process.

The speculation provided the agency with a legal rationale for constraining internet services to its predetermined model: Unless internet services fit the agency’s paradigm, platform innovation would not occur, and there would be no pressure on broadband networks to improve.

Meticulous measurements of broadband performance on the part of regulators and private firms confirm that speeds have improved at a 35 percent annual rate for the past decade. Contrary to the virtuous circle hypothesis, web speed—the time it takes for webpages to load—has only modestly progressed overall and has even regressed since the FCC’s 2015 action. Web performance has not been subject to the level of scrutiny focused in broadband platforms. This is due in part to the difficulty in measuring web performance. But the FCC has covertly politicized performance measurements. Its Measuring Broadband America (MBA) reports examine both webpage load times and broadband speeds but fail to analyze web data properly.

The load time of webpages does not improve over broadband networks faster than 12–15 Mbps. Early MBA reports reported this fact correctly, but those issued after the FCC redefined “broadband” to 25 Mbps have claimed a threshold value of 25 Mbps even though the underlying data have not changed.

The emphasis on one facet of internet performance, such as last mile broadband networks, tends to minimize other factors that may be more important to the user, such as the performance impact of tracking networks, browsers, webpage design, and web server performance. In addition, relying on active measurement tools creates opportunities for gaming the system that are not possible in passive systems that merely observe application and network events in real time.

This paper explores opportunities for developing performance tools more responsive to the broader social goal of better end-to-end internet performance across the broad span of applications. It finds that systems for capturing passive measurements and sharing them among ISPs, web developers, and other responsible parties may be useful for accelerating the web experience.

The performance of websites over time is a neglected facet of internet measurement that deserves more attention. In an era of increasing internet consolidation, smaller sites and platforms are squeezed by larger competitors able to invest in private infrastructure. Better insight into web performance and increased flexibility in contracts between content platforms and broadband platforms may mitigate investment inequality effects.

Bottom line: users don’t get the full benefit of faster broadband because ad networks eat it up.