The Skype outage

Did you know...?

LWN.net is a subscriber-supported publication; we rely on subscribers to keep the entire operation going. Please help out by buying a subscription and keeping LWN on the net.

A recent outage at Voice over IP (VoIP) provider Skype has caused quite a stir. For nearly two days, users of the VoIP software could not make calls, which set off a storm of blog postings wondering about the cause. Skype released an official explanation that did not ring true to some, leading to further speculation.

Sometime early Thursday, 16 August, Skype users could no longer authenticate and connect to the network. On Friday, right in the middle of the outage, a posting to Bugtraq purported to have information about the vulnerability that was being exploited to cause the outage. Skype has since categorically denied that any attack was responsible, but suspicions persist that the denial-of-service (DoS) vulnerability reported was actually responsible for the outage.

On Monday, Skype posted the following to their Heartbeat blog:

The new message provided more details, but still remained mute on one of the central puzzles: why did updates on Tuesday cause an outage starting on Thursday? While they acknowledge a bug in their software, there is also no mention of how the situation was resolved, presumably through an automatic update of their own. Overall, the explanations are fairly thin on technical detail which allows others to conjecture to try and fill in the holes.

There are many millions of Skype users – the software is available for Windows, OS X and x86 Linux – for the no-cost PC-to-PC calling as well as the other services that Skype does charge for. Hopefully the free users are not depending on the service, but there are companies which use Skype exclusively; an outage for two weekdays must have been rather painful. Certainly the landline and cellular phone companies have had their problems along the way, but those tend to be regional rather than worldwide.

All software even minimally more complicated than "hello world" has bugs, and those bugs will be triggered in surprising ways. Taking the Skype "perfect storm" explanation at face value, it is nearly amazing that millions of reboots could result in a network storm so severe that it would take two days to resolve. Somehow, in the interface between the Skype's centralized authentication and their P2P routing code, things went horribly awry. It does, however, give one pause about the power of the near-monoculture in desktop operating systems.

It is hard, but not completely impossible, to imagine a similar scenario for Linux boxes. To start with, it is uncommon that a software upgrade requires a reboot. Within the Linux user community, there is a wide range of kernel versions running, so even if there were a critical security fix that required "all" Linux kernels to be upgraded, it would not be very synchronized – the distributions tend to have different response times. This is a bit of a double-edged sword, of course, those varying response times could leave a hole that a worm or attacker could exploit. But, because Linux boxes are controlled by their owners rather than by their distribution provider, synchronized reboots are probably not a major cause for concern.

Beyond monocultural issues, there is the question of how a P2P system can be taken down by the lack of a centralized resource, in this case credentials from an authentication server. That provides a single point of failure to what is supposed to be a robust architecture, resistant to exactly those kinds of problems. There are also those who wonder if the outage was caused by an "upgrade" mandated by the US government so that they can more easily monitor Skype calls.

Skype is proprietary and closed source; there is no easy way to determine whether the problem has been fixed, or even whether the problem is being accurately described. If Skype decides, or is forced, to change their software to be more easily monitored, it will be hard to detect. It might look an awful lot like a multi-day outage that clears up somewhat mysteriously. Trusting closed source software for vital communications is not the best of plans, at least when there are alternatives.

Free software would not necessarily avoid these kinds of problems, but a completely decentralized network with multiple clients sharing a protocol, but little else, would certainly be more resistant to this kind of outage. More importantly, it would also be more transparent. Over time, projects like openwengo, Linphone, Asterisk and others can hopefully provide those benefits to a larger audience

Note: you can avoid this step in the future by logging into your LWN account.