On Thursday, February 03, 2011 17:01:27 Pablo Neira Ayuso wrote: > On 03/02/11 16:42, Pablo Neira Ayuso wrote: > > On 03/02/11 15:23, Hans Schillstrom wrote: > >> On Thu, 2011-02-03 at 14:51 +0100, Pablo Neira Ayuso wrote: > >>> On 03/02/11 14:34, Hans Schillstrom wrote: > >>> this assumption is not valid in NAT handlings. > >> > >> That's true, because I want to avoid conntrack > >> > >>> If you want consistent hashing with NAT handlings you'll have to make > >>> this stateful and use the conntrack source and reply directions of the > >>> original tuples (thus making it stateful). That may be a problem because > >>> some people may want to use this without enabling connection tracking. > >> > >> What about a compilation switch or a sysctl ? > > > > or better some option for iptables. > > Hm, this is actually not straight forward to implement, you'll have to > use hook functions to avoid the module dependencies with conntrack and > that's pretty annoying. > > I don't come up with a good solution for this. A configuration switch might be OK. > > >>> Are you using this for (uplink) load balancing? > >> > >> Actually in both ways > >> - in front of a bunch of ipvs > > to make some preliminary load-sharing between the load balancers? Yes that's right and in the payloads send the return traffic in the same path. > > >> - and in the payloads for outgoing traffic. > > and then to select the uplink, right? > Yes. It also has the same role for cluster originated traffic to spread the load over multiple interfaces, and catch the return traffic. -- To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html