Youve got firewalls and VPN concentrators at the very outside. Youve got intrusion detection and prevention, virus scanners, e-mail security devices, and boxes to do network access control and content filtering. Of course you might have outsourced some services, such as e-mail hygiene. But others really need to be done at your actual network perimeter.
And all these boxes run in-line, creating a tenuous physical architecture, potential points of failure and multiple potential performance bottlenecks, but, as Arlo Guthrie said, thats not what I came here to tell you about.
I came to talk about the toughest job in the whole perimeter: the intrusion prevention system, or IPS. It turns out that the IPS not only has the most difficult job performed at the network perimeter, its not generally taken seriously.
The IPS at the network perimeter evolved out of the IDS (intrusion detection system), which scans network traffic looking for signs of attack and reports them. IDSes developed the reputation of bombarding administrators with reports, very few of which really needed to be dealt with.
IPSes, like IDSes, are driven by attack signatures for which they scan the traffic. An IPS goes a step further: It not only scans for attacks, but also attempts to block them. Theres another type of IPS—a host-based IPS—which runs on a computer and attempts to block attacks aimed at that specific computer.
Unsurprisingly, network IPSes suffer from all the problems of network IDSes: Depending on how you tune them, they are prone to false positives and have the potential to slow all network traffic. Host-based IPSes have a significant advantage: Since they run on the computer theyre protecting, they have the ability to monitor the state of that system and the context of the attack.
Both network and host-based IDSes have the ability to detect specific attacks that have a specific signature. Both can, to some degree, detect some attacks generically, such as stack-based buffer overflows. But the host-based IPS can look at the state of registers and processes into which that potential buffer overflow is traveling. It has more information from which to make intelligent decisions.
Its theoretically possible to make a network IPS that tracks context in the same way for each of the systems its monitoring. Imagine the load on such a system, and imagine having to run it in-line.
This is the trade-off that dooms the IPS: In order to monitor enough context to make an intelligent decision about attacks, an IPS would have to consume resources and time such that it would be a cost and performance burden. If you tune it down to the point where it doesnt impose such a burden, it wont have context and youll miss attacks, or youll get false positives, or possibly both.
This is why many administrators turn off blocking, effectively turning their IPS devices into IDS devices. Maybe they have the idea that theyll forensically examine the logs over time and see if theyre trustworthy enough to turn on blocking. Do you really want to spend time on this or would you rather go to the ball game? Youll find me in section 205.
A network-based IPS makes for great theory, but I wouldnt trust one that wasnt severely constrained to well-proven detections. Its just too hard a job for an in-line device.
Security Center Editor Larry Seltzer has worked in and written about the computer industry since 1983.
More from Larry Seltzer