On Wed, Jun 22, 2011 at 3:57 PM, Juliusz Chroboczek <[email protected]> wrote: >> I believe this is indeed working around an old kernel bug. We needed to >> make this work with some old 2.4.x kernel. A quick test on 2.6.32 >> suggests that adding a two routes with the same metric and different >> gateways works just fine now. > > I may be wrong, but I believe that this is due to IP_ROUTE_MULTIPATH > being enabled in your kernel, and has nothing to do with the kernel's > vintage.
Interesting. I did go to the trouble of coding up a solution that tried to add then delete, and fell back to the old behavior if it had trouble... and I just survived a thunderstorm with it. Thus far, no kittens eaten... > Dave, please note that the window when there's no route is very short -- > it lasts just a few microseconds --, so you'll typically lose just > between 0 and 1 packets on a route change. It seemed considerably higher than that on a busy router. My assumption was that the process could be rescheduled between the calls to netlink, while considerably more work was getting done in a softirq... Or (as usuall) I'm battling another bug. I'm getting closer to where I can finally get insight into the behavior of huge streams on tiny routers by the day... That said, since I just coded up the solution I'll try to time the transaction somehow to gain some better insight (without skewing the test)... > I agree that it should be > fixed, and using Linux' ability to do multipath is a good workaround, as > long as it doesn't break babeld on non-multipath kernels. I do like the idea of using multipath to a greater extent. > > -- Juliusz > -- Dave Täht SKYPE: davetaht US Tel: 1-239-829-5608 http://the-edge.blogspot.com _______________________________________________ Babel-users mailing list [email protected] http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/babel-users
