Data is at its most valuable just after it's created, and vulnerability data in particular has a short half-life, says Packetloop's Michael Baker.
And that, he told delegates to AusCERT 2013, lay behind the open sourcer’s increasing focus on two things: using the disciplines of big data—particularly Map Reduce—to help lift the lid on attack traffic, and to do that ever-closer to real time (replacing Map Reduce with Streams).
Only with the advent of big data, Baker claims, has it become clear that our use and application of Bayesian techniques has been limited to date: “When you fire terabytes through the network detection you discover how limited that is.”
For example, he harked back to the “Apache-killer” CVE-2011-3192, in which nearly trivial signature difference between Snort and Metasploit illustrated how poorly even similar signatures are identified by Bayesian software.
“We're trying to build an industry on log file correlation: I don't get it,” he said.
“No matter how fast Map Reduce on Hadoop can identify patterns that might identify traffic—and no matter how frequently it's scheduled to run an update—it only ever provides information about the past.”
“Map Reduce is not pretty—but it's like a Clydesdale, it can carry the load.”
Hence the decision to implement a cloud-based PacketLoop rather than relying on post-hoc analysis. The aim is to create both an architecture and scale that is suitable for detecting attacks in real time.
AusCERT 2013 : Day 1 Coverage
AusCERT 2013: Day 2 Coverage