The other day I was blogging on the Damballa site about the principles behind dynamic reputation systems - Building A Dynamic Reputation - and trying to answer a question that came up over whether dynamic reputation systems can replace IPS.
You'll find some comments on the other blog, but I wanted to add some more thoughts here - based upon some thoughts shared by others on the topic.
I guess the issue lying at the heart of the question is whether, by implementing a blocking (or filtering) policy based upon the findings/classification of a dynamic reputation system, you'd be gaining better protection than having implemented a stand alone IPS.
To issues come in to play in the the decision - How "complete" is the dynamic reputation system? and How "reliable" is the IPS?
As I said in the original posting - advanced dynamic reputation systems have been coming along in leaps and bounds. We're not talking about some static blacklist here and neither are we limiting things to classic IP reputation systems that deal with one threat category at a time. Instead we're talking about systems that take as inputs dozens of vetted threat detection and classification lists, realtime feeds of streaming DNS/Domain/Netflow/Registration/SpamTrap/Sinkhole/etc. data and advanced machine learning algorithms.
From experience (and empirical evidence), blocking the things that a dynamic reputation system says is bad or very suspicious at the network perimeter appears to out perform IPS - if the count of victim machines is anything to go by.
One of the key failings of IPS is that its reputation is better than its performance. What I mean by that is an IPS is limited to its signatures/algorithms for detecing know threat profiles and exploit techniques. These are not all encompasing - and you'll normally only fine the first "in-the-wild" exploit for a vulnerability covered (or exploits that get used by popular commercial hacking tools and IPS testing agencies) - rather than all the obfuscation and evasion techniques. You may remember the blog I did a little while about the commercial exploit testing services used by the badguys - such as Virtest.com.
So, here's my thinking. It's better to block known bad and provable dangerous/suspicious servers (independent or restricted to a particular protocol - depending upon your tolerance for pain) than on a hope that your IPS is going to stop some (hopefully) past-seen permutation of a particular exploit being served by the attacking server.
Some may argue that you're still at risk of servers that are unkown to a dynamic reputation system. Are you though? Think of it this way. You have a dynamic reputation system that is taking live datafeeds etc (as described above) for the entire Internet. If a server (or service) has never been seen and doesn't have a reputational score - then it's already suspicious and could probably be blocked for the timebeing.
Defense in depth is still a good option though!