On a good day, UUnet experiences packet losses of 0.04%. On Thursday, the packet losses peaked at 22%. When that happens, the network is clogged all day with crisscrossing TCP/IP messages looking for confirmation that never comes. That translates into very little real data getting across and outright application downtime for many customers.
Service was back up and running around 4 p.m. EDT. According to a company that has customers in common with WorldCom, engineers at UUnet spent Wednesday night upgrading software on thousands of perimeter routers that communicate with vendor partners and WorldCom ISP customers. They didn't know about an operational glitch with the upgrade when they turned the routers on at 7 a.m. Thanks to that glitch, UUnet's availability for nine hours went from 95% to less than 70%.
Keynote Systems, Matrix NetSystems, and Sockeye Networks monitor app or Internet performance and spent the day steering customers to other backbones and otherwise keeping them informed. WorldCom wouldn't comment on the damage.
"We've never seen anything like this in the history of the Internet," says Tom Ohlsson, VP of marketing and business development at Matrix NetSystems. "Any company that only hosts with one ISP learned a lesson today. And they want to host with carriers who use different backbones and don't have interlocking agreements."