nanog mailing list archives

Re: ACLs vs. full firewalls


From: Roland Dobbins <rdobbins () cisco com>
Date: Wed, 8 Apr 2009 08:28:47 +0800


On Apr 8, 2009, at 4:05 AM, Michael Helmeste wrote:

However, I wanted to get other opinions of what packet filtering solutions people use in the border and in the
core, and why.


Stateless ACLs in hardware at the edge are important both for infrastructure self-protection (i.e., iACLs) and for policy enforcement of the type you indicate. As others on this thread have pointed out, do understand your platform characteristics and craft your ACLs accordingly.

Stateful - i.e., context-aware bidirectional - filtering via a firewall makes sense in situations in which a) the nodes 'behind' the firewall aren't typically operating as servers and/or b) the bidirectional communications patterns which should be observed are well-known, and in which the participation of hosts is under the control/influence of the network operator. For example, in front of a corporate LAN, or between the tiers of a multi-tiered application, one can craft quite specific stateful inspection rules which can be used to explicitly allow and disallow certain types of traffic.

For front-end, publicly-accessible conventional servers, stateful inspection may not add as much value, as basically every connection which comes into those servers is unsolicited (i.e., no existing stateful communications context against which to measure pass/fail decisions); this is where high-speed stateless ACLs, coupled with host OS/app/service hardening play a key role. It's very important to avoid the instantiation of unnecessary state in front of public-facing assets, as DDoS attacks are essentially attacks against capacity and against state.

One should also look into implementing DDoS mitigation techniques such as S/RTBH, in conjunction with the chosen policy-enforcement regime.

-----------------------------------------------------------------------
Roland Dobbins <rdobbins () cisco com> // +852.9133.2844 mobile

  Our dreams are still big; it's just the future that got small.

                   -- Jason Scott



Current thread: