Cloudflare: Internet traffic was down 3.5 percent due to CenturyLink outage
A major outage at the American provider CenturyLink / Level (3) caused worldwide internet problems on Sunday. According to Cloudflare, this traffic fell by 3.5 percent. The cause can be traced to it border gateway protocol.
On Sunday, traffic from the major American provider CenturyLink / Level (3) almost came to a standstill due to a disruption. The consequences were felt worldwide and international websites and services were temporarily unavailable. CenturyLink provides network services to enterprise customers worldwide. The company has not yet provided a detailed explanation of the outage itself, but Cloudflare already describes what happened and how the outage affected its customers.
On Sunday morning, Cloudflare received a large number of ‘522 errors’, as a sign that there was a problem with the connection from its network to places where customers host their data. Cloudflare’s systems then began redirecting traffic from CenturyLink / Level (3), the source of the problems, to alternative network providers such as Cogent, NTT, GTT, Telia and Tata. A graph of tier-1 networks connected to Cloudflare shows Centurylink traffic, shown in red, virtually coming to a standstill.
For some of the Cloudflare customers, it took a while for the connection to be re-established, because the systems have to take the capacity of the networks into account when diverting. Moreover, it turned out that some of the customers were only connected to the internet via CenturyLink and that provider did not honor some of the redirects.
The cause of the malfunction was instability of border gateway protocolroutes in the CenturyLink backbone, especially at a flowspecupdate, according to a CenturyLink announcement. The border gateway protocol is the routing protocol that regulates traffic between providers and flowspec is an extension to this with which firewall rules can be distributed within networks. It is not yet known what went wrong with the flow spec update and why it took a relatively long time to fix the problem.