The Internet Report returns this week after Thanksgiving, although this installment is probably more correctly described as a “fortnightly pulse” rather than the usual “weekly pulse,” since we’re taking into account an extended set of numbers.
That said, outage numbers were substantially down worldwide through the Thanksgiving holiday period. There were 25% fewer outages in total, with North America in particular down 44% over the previous week. North America also accounted for only 36% of all outages we observed, whereas typically it would represent 46-48%.
UCaaS outages (which are commonly associated with remote work) dropped a massive 117% globally and 70% in North America.
The week following the Thanksgiving holiday saw another marginal decrease, with total global outages dropping a marginal 1%, and outages observed domestically increasing by 11% compared to the previous week. Even with this increase, outages were (on average) 36% lower than what we had observed in the weeks leading up to the holiday.
The slowdown in outage numbers in North America is to be expected. With people spending time with family, businesses go into a shutdown mode. In prior weeks, I pointed to an uptick in outages that bore the hallmarks of maintenance works.
My hypothesis is that most network operators fast-tracked maintenance to occur before the holiday period, and therefore we saw fewer outages in the holidays because the maintenance work was already done.
But that is likely to be only part of the story.
An equally (if not more) important factor that likely caused outage numbers to fall around Thanksgiving this year was all the work that was performed back in 2020 to prepare for the first pandemic-era Thanksgiving period and the traditional Black Friday and Cyber Monday sales. One of the reasons we saw less disruption this year could be because engineers had previously built extra capacity into their systems to cope with the expected rush to online, with shelter in place restrictions reducing the potential of in-person physical sales opportunities. It’s likely that 2021 was their payoff year, with engineers enjoying some personal downtime while their systems stayed up.
There may also have been other factors at play. This year, Black Friday wasn’t a single day or weekend, but instead, it was spread across the full month of November. Outside of North America, some Black Friday sales did not wrap up until December 5. Spreading the sales across a number of weeks altered the online traffic profile of the sale from a concentrated 24–72-hour spike to a more load-balanced event. It avoided a single aggregate traffic event. Even though we know websites have been re-engineered to scale, they didn’t need to because the traffic aimed at online retailers was more spread out and any peaks were far less severe.
Some shoppers may not have waited for this year’s sales, either. Global supply chain issues have been well-documented, impacting product inventory, order fulfillment and last-mile delivery. Knowing that shipping volumes were already elevated, shoppers were encouraged to start earlier this year to provide an extra time cushion to receive their online orders. Others may be choosing to shop locally from existing retail inventories to avoid any potential issues with online fulfillment altogether.
One thing worth noting is that while outages were down substantially in the period, we did observe slowdown error indications on e-commerce sites. For example, when performing a lookup on item inventory before placing an order. Internet users will be aware of the “eight-second rule,” which defines people’s patience for a page to load before they get bored and go elsewhere. That was revised down to two seconds by 2018.
During this year’s Black Friday sales, we observed some page load times in excess of eight seconds, and yet no social media complaints. Perhaps because of widespread understanding of the pressures that retailers, infrastructure, supply chains and the people around them are under, shoppers appeared to be more tolerant. They were prepared to put up with some performance degradation to transact and get a bargain.
More detailed research is needed to determine whether this tolerance is temporary, and the extent to which the conventional wisdom around Internet impact on attention spans has been challenged.