The last months we’ve been working hard on improving Suricata. So hard actually, that we’ve drifted a bit from our original goal of doing a 1.0.3 “maintenance” release. Instead, the new release will be 1.1beta1. The change to 1.1 is to indicate the large number of changes, the beta1 is to … indicate the large number of changes 🙂
As you may know, Will Metcalf moved on to join Qualys. A significant loss to our project as Will was one of our founding members and is hard to replace in his role as QA lead. Not having a full time QA person on the team right now is a reason for us to decide we’re in need of a beta cycle for the next release.
So… what kind of improvements are we talking about?
- Improved parsers, especially the DCERPC parser.
- New keyword support: http_raw_header, http_stat_msg, http_stat_code.
- Much improved fast_pattern support, including for http_uri, http_client_body, http_header, http_raw_header.
- A new default pattern matcher, Aho-Corasick based, that uses much less memory.
- Lots of small performance updates, including SSE3, SSE4.1 and SSE4.2 optimizations.
- The signature bitmask prefiltering I wrote about before.
- We support the reference.config supplied by ET(pro) and VRT now.
Lots of mention of performance in this list. Did it improve? Yes! As some of you may have read, Npulse has demonstrated 10 Gbps IDS support for Suricata using Napatech (PDF) hardware support. This was on fast hardware, but nothing outrageous. To be honest, I didn’t expect to get there yet. But they did it. Based on a slightly modified Suricata 1.0.1 and about 7k signatures. Our own testing has shown that the code has improved quite a bit since then: ranging from 25% to 67% more packets per second throughput. Btw, native Napatech support is expected to go into our code base sometime in the next few weeks.
We have two major areas where we want more improvement. The first is the inline mode. Due to Suricata’s HTTP and other protocol parsers working statefully on top of the stream reassembly engine, currently all work is done on ack’d data. This means dropping attacks based on keywords such as http_uri is hard. We’re planning a number of changes to the stream engine to address this. More on that in a future post. The second area is the rule language. At this point we still miss a number of keywords to properly support mostly VRT signatures. Keywords like file_data.
The current git master is pretty much what Suricata 1.1beta1 is going to be. The actual release is planned for next week, probably Tuesday or Wednesday. If you can, help us out by trying it and report any issue to us!