A wave of brief internet outages hit the websites and apps of dozens of financial institutions, airlines, and other companies during the week of June 14. Experts say it highlights the web’s vulnerability to shutdowns, and its growing reliance on a chain of servers called Content Distribution Networks (CDNs), which are responsible for the outages. CDNs have become increasingly popular in recent years, Olaf Kolkman, a principal at the Internet Society, a nonprofit that advocates for an open internet, said in an email interview. “But the big drawback is that if something goes wrong in a CDN central configuration system, or there is a cybersecurity issue, then a lot of content goes down,” Kolkman added.
Handy, but Problematic?
Most of the websites affected by the internet outage are served by the company Fastly, which is among the world’s biggest CDN providers. Another CDN, Akamai, said about 500 of its customers were affected after a software bug. “Many of the approximately 500 customers using this service were automatically rerouted, which restored operations within a few minutes,” the company said in a statement on its website. “The large majority of the remaining customers manually rerouted shortly thereafter.” CDNs are gaining more traffic because they allow local distribution of data instead of sending it through undersea cables. “So if you host popular content, it’s cheaper to install a server in a few 100 cities so that all those Internet users can get content from nearby, vs having to pay for the transit of content that needs to travel long haul,” Kolkman said. CDNs also offer fast connection speeds and resilience against cyberattacks, Kolkman explained. “However, CDNs are a distributed infrastructure and managed by one entity, meaning a mistake or attack on the backend infrastructure that configures these CDNs may impact all the distribution points,” she added. “And because these CDNs typically have many customers, there will be a lot of content that ‘disappears’ or isn’t accessible—which is exactly what happened with the recent Fastly outage.”
Internet Vulnerabilities Abound
CDNs aren’t the only reason the internet is vulnerable. The basic structure of the web lends itself to outages, Ataollah Etemadi, the head of web hosting company DivisionX, explained in an email interview. That’s because the web is controlled by software whose specifications are freely available, he pointed out. “On the plus side, that is great because devices can “speak” the same language,” he said. “On the minus side, it means that if there is a bug or issue, it can affect millions if not billions of devices. It has always been well known that the internet is the most hostile environment possible for code.” Engineers often have to spend painful hours hunting through logs and dashboards to find the root causes of outages. To solve internet software issues, some providers are turning to machine learning systems. Zebrium, for example, offers software that learns to uncover problems automatically. Outages often occur not because of major widespread issues, but rather because of some kind of subtle software failure, Gavin Cohen, a vice president at Zebrium, said in an email interview. “Every environment is different, and there [are] almost an infinite number of possible failure modes,” Cohen added. “When a problem does occur, it’s key that a company gets to the bottom of it ASAP. Instead of humans having to troubleshoot manually, machine learning can do this almost instantly and more reliably.” Etemadi doesn’t think we’ll ever be able to prevent internet outages completely. “The internet is made up of software, and software has bugs,” he said. “The software can be hacked. You can only plan for and mitigate it.”
