CERT Stats Under Fire

Linux supporters have roundly criticized a recent report from the United States Computer Emergency Readiness Team (CERT), which reported that during 2005, Linux and Unix combined had 2,328 vulnerabilities, compared with 812 vulnerabilities for Microsoft Windows.

Linux practitioners say the counts are skewed because they count the same vulnerability each time it appeared last year in any given Linux distribution. By doing this, they say, one bug could actually show up in the list dozens of times, depending on the number of Linux variants it appeared in. The CERT stats also appear to include problems with scripting languages such as PHP or even applications that are not part of the core Linux operating system but instead are used with it.

Another complaint about the CERT numbers, aside from lumping Unix and Linux into the same bucket, is a behavior specific to the open-source community. There might well be multiple reports for the same bug, says David Humphrey, a senior technology adviser for Ekaru, a Westford, Mass.-based consultancy. In the open-source world, a bug report is issued anytime something is originally discovered, of course, but then at each stage of the fix there is usually another report issued. "When it's worked on and some progress is made, there's another report, and so on," he says. This contrasts with Microsoft's approach, he says, which typically is to report a bug only after it's been resolved, and only once.

“I don't think that you can accurately draw conclusions from the CERT report," says Dave Rosenberg, senior analyst at the Open Source Development Labs, a vendor consortium that helps maintain the core Linux kernel. "This report doesn't provide any beneficial information for CIOs or IT staff making security decisions."

As of press time, CERT spokesmen did not return numerous phone calls seeking comment on the organization's counting methodology. But Alan Paller, director of research at the SANS Institute, defended CERT as having a "rigorous, consistent process. The fault is not with CERT; the fault is that there are so many different distributions of Linux." Still, even he conceded that the CERT statistics are not really usable except as a jumping off point to discuss security issues.