Wednesday, April 8, 2020

Fastly measures traffic spikes vs download performance

Fastly published traffic analysis that delves into regional trends for key U.S. states, as well as the countries of France, Italy, Japan, Spain, and the UK, that are some of the most impacted by the COVID-19 pandemic (as of the end of March 2020). The overall conclusion is that the Internet is holding up to the surging traffic volumes resulting from home quarantines.

“Overall, the internet is in good health,” said Fastly’s Chief Architect and Founder, Artur Bergman. “This is partly due to the regionality of these trends, but modern websites and applications are also better able to adapt to changing internet conditions. We are seeing the internet bring people together, whether for work, entertainment, or to get in touch with family and friends. And while there’s more traffic than in previous months, the internet is resilient.”

For analysis of regional traffic, Fastly used the following metrics, which were gathered from sampled TCP connection stats (tcp_info) at connection close time as part of Fastly’s fleet-wide performance monitoring infrastructure:

Traffic represents the average number of data bytes per day that were delivered to various geographies from all of our servers (acked_bytes field in tcp_info).
Download speed represents the delivery rate reported by TCP over all connections from a region, averaged over a day (delivery_rate field in tcp_info). While the reported values are averages, various percentiles for download speeds all show similar trends and ratios to the average.
For global analysis of verticals, Fastly measured average requests per second (RPS) recorded at all our servers. RPS is a count of requests received at our serves from end users every second.

Some highlights


  • One-month traffic trends: Over one month, from respective traffic baselines, all regions analyzed reported increases in traffic with decreases in download speed, with the exceptions of Japan and California.
  • Most notably, Italy saw a 109.3% increase in traffic, with a 35.4% decrease in download speed.
  • Japan observed a 31.5% increase in traffic, but, markedly differentiated from other analyzed countries experiencing the impacts of COVID-19’s pandemic in their communities, it saw a 9.7% increase in download speed. Japan’s internet infrastructure seems to have absorbed this increased traffic.
  • In the U.S., New York and New Jersey saw a 44.6% increase in traffic, but were able to weather it with a relatively modest 5.5% decrease in download speed.
  • California saw a 46.5% increase in traffic, with a 1.2% increase in download speed. Similar to Japan, California’s internet infrastructure appears to have absorbed this increased traffic.
  • School closures and regional stay-at-home orders: In France, Fastly observed a 45.4% increase in traffic shortly after its government announced school closures on March 12. This increase in traffic did not show a noticeable reduction in download speed, but on March 17, when schools closed and the national lockdown went into effect, download speed decreased by 20.6%. In Italy, traffic increased by 47% around February 19, after public information about lockdowns of various parts of Northern Italy began. After school closures go into effect on February 24, download speeds fall by about 9.2% between February 18 and February 24. In Japan, traffic jumps by 38% and speed decreases by 9.2% between February 25 and February 29, when school closure announcements began. School closures in Spain on March 12 mark the beginning of traffic increases of 26.2%, while there is no speed reduction.
  • User experience: Across European nations, major streaming and VOD providers announced they would reduce the default bitrates of their video streams around March 19 in order to prevent the internet from acute strain. When analyzing traffic and download speed in France, Italy, and Spain, the reductions do not seem to have led to significant improvements in internet quality, but they may have helped keep it from deteriorating further. Additionally, beyond Europe, people who have had lower speed internet connections before coronavirus-related developments may currently be suffering from greater degradation than those who had higher speed broadband connections. Someone with a high-quality, 100Mbps internet connection might not even notice a download speed reduction of 35% — a typical 1080P video stream requires just 6Mbps and fits well within the resulting 65Mbps download speed. However someone with a lower-quality connection to the internet, such as 10Mbps, is likely to notice this quality degradation: their typical 1080P movie might now automatically adjust down to 720P to fit within the resulting 6.5Mbps.
  • From these findings, we learn that speed decreases may be more aligned with population shifts toward more internet use from homes, rather than with a traffic increase alone. In general, most regions saw their download speeds stabilize once school closures and lockdowns were fully enacted, when most of the population became homebound. For most regions, this happened mid-to-late March.


Fastly also analyzed industry-specific internet activity by comparing average requests per second (RPS), week-over-week, between two sets of dates: January 6, 2020, and February 16, 2020; and February 16, 2020, to March 29, 2020. The first set of dates represent what Fastly considers to be activity attributed to organic growth, as mainstream attention on the pandemic had not yet picked up to its current pace. The second set of dates occur during a much more dynamic era of coronavirus-related developments, and was instructive in helping understand how human behavior may be shifting in response to COVID-19.

During those time frames for each of the below verticals, Fastly observed the following traffic patterns:

Streaming: From February 16 to March 29, streaming observed an increase in average RPS week-over-week by 29.6%. Similarly to gaming, this notable increase could reflect increased interest in streaming media content during lockdowns and social distancing.
News and digital publishing: Of all verticals analyzed, these brands saw the biggest increase in average RPS week-over-week from February 16 to March 29 at 70.16%. The elevated activity may correlate to increasing newsroom attention on COVID-19-related content as the first quarter of the year progressed.
Social Media: Social media platforms saw a 40.88% increase in average RPS week-over-week from February 16 to March 29. The greater increases in activity might represent a window into one way people are remaining connected with each other on various social media platforms.
GIFs/memes: Brands that help end users create and share things like memes and GIFs observed a 30.28% increase in average RPS week-over-week from February 16 to March 29. The unusually high increase in activity may indicate that, in reaction to disruptions to daily life, some consumers have turned to humor to get through difficult times and connect with others.
Gaming: From February 16 to March 29, gaming observed an increase in average RPS week-over-week by 28.54%. This could indicate that people are playing more virtual and online games as many communities look for ways to stay engaged while sheltering in place.
EdTech: From February 16 to March 29, edtech observed a noticeably sharp increase in average RPS week-over-week by 34.55%. This could indicate that edtech platforms are indeed growing their usage as more school-age children are educated from home.

https://www.fastly.com/blog/how-covid-19-is-affecting-internet-performance

See also