David Lamparter has two patches that fix prefix-list algorithmic issues and greatly speed up the run times of prefix-lists:
http://patchwork.quagga.net/patch/1147/ http://patchwork.quagga.net/patch/1146/ Looks like we should get them into the tree, sooner rather than later. donald On Wed, Oct 21, 2015 at 2:49 PM, Nick Hilliard <[email protected]> wrote: > On 21/10/2015 17:23, Daniel Walton wrote: > > What is the ballpark maximum number of peers a route-server running > quagga > > will have? I tested with 300 peers and bgpd takes ~5% of the CPU (2.4Ghz > > x86)...the peers are nice and stable > > this depends on prefix distribution. > > the problem is peer startup / shutdown and larger scale flaps where you > have lots of updates. At INEX, we have ~70 peers, but one of them is > injecting ~30k prefixes. Each inbound policy was attached to a > prefix-list, and each export policy had a route-map with a community-list > attached. This config caused cascading failure caused by resource > starvation when that large peer connected, which could only be remedied by > starting up the large peer first, then manually starting up groups of > smaller peers. We ended up having to migrate to bird because quagga > couldn't handle this. > > Before then, we had anywhere between 2k prefixes and 8k prefixes depending > on peoples' whims, and even at that stage we were seeing piles of slow > thread warnings. > > Nick > > > _______________________________________________ > Quagga-dev mailing list > [email protected] > https://lists.quagga.net/mailman/listinfo/quagga-dev >
_______________________________________________ Quagga-dev mailing list [email protected] https://lists.quagga.net/mailman/listinfo/quagga-dev
