On 01/03/2014 01:45 AM, Gert Doering wrote:
If there is no attack, rate-limiting won't kick in. This wasn't about "rate-limit to 10pps" but to "less than 10Mbit/s" - which you just won't see under normal circumstances, if you exclude well-known local recursives plus well-known remote recursive from the rate-limiting.


Exactly my point and the reason why I brought this thread up.

We had another such incident tonight, throwing over 700mbps at us. My router stood up, my filter kicked in, the network continued functioning. Someone here pointed out a flaw in my original access-list which didn't catch everything I thought it did and now with that fixed under this latest incident I think the results pretty clearly are that this seems to be a reasonable step. The open question now for me is what level of granularity do I really want to implement?

I have been looking at MCQ documentation to learn how I might implement a better policy than one-queue-or-rate-limit-fits-all and as it applies to my situation (broadband provider / pppoe subscribers). Some of the comments here are helpful - I agree it's likely I could have a short list of known good resolvers that don't need to be rate limited, and of the unknown ones its likely I could have perhaps a per-user qos to ensure a subscriber has some minimum guaranteed bandwidth for dns to any destination. Globally I also of course would likely set a global catch-all to rate limit traffic not caught by the customer qos limits to something like 15mbps total. What I don't know here is what exactly is supported on my router (7201 with 12.2(33)-sre7) nor what the net effect on my router is likely to be for > 1000 pppoe sessions if I did something like this. While I was at it, if I had to have some kind of qos set up for users, prioritizing dns is likely a good step in any event but Im sure there are others as well. Perhaps I might like to also prioritize sip over http over unclassified bulk traffic?

I just dont have a good handle on how to attack this or what these traffic classes / policies would actually look like just yet. Does anyone else do this that could post a config snippet?

Mike-


_______________________________________________
cisco-nsp mailing list  [email protected]
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

Reply via email to