[Xorp-users] Problem using xorp on real internet ...

Marko Zec zec at icir.org
Mon Jul 2 05:18:21 PDT 2007


On Monday 02 July 2007 12:32, Benjamin Sonntag wrote:
> Hi,
>
> I'm Benjamin Sonntag, from Gitoyen.net, a french  network operator (I
> made an up-to-date package for xorp on debian)
>
> We are currently using quagga, but we have some problems with it (eg
> bgp flapping during transit disconnection)
>
> We are testing xorp and we hope we will be able to use it in
> production, but a (quite huge) problem appeared during our tests :
>
> We put our tests notes & results here (maybe in french, sorry)
> https://wiki.gitoyen.net/xorp/
>
> our second test is when xorp receive all the Internet routes from a
> bgp transit operator (called bsd1 here). There is about 200K routes.
>
> The configuration used is here :
> https://wiki.gitoyen.net/xorp/config2.deb1.boot
> and the bgp peer is an openbgpd whose config is here
> https://wiki.gitoyen.net/xorp/config2.bsd1.boot
>
> The point is that XORP took about 35 minutes (on this mono-xeon box)
> to compute and add the routes to the fib ! During this time,
> xorp_bgpd and other processes took all the cpu cycles.
>
> Is this behavior normal ? Is Xorp bound to be used by network
> operator using bgp or does it need some (big) optimizations ?
>
> Is it something that can be done to make this far more lightweight ?
> or should we use openbgpd instead for this kind of usage ?

Hi,

you might try to configure and build XORP with a few optimization flags 
enabled, as described at the very bottom of the BUILD_NOTES file.  The 
--enable-optimize and --disable-debug flags in particular should yield 
significant improvements, while the others are very much compiler 
specific but might be worth playing with.  Nevertheless, even with 
compile-time optimizations enabled, XORP currently lags behind Quagga 
and OpenBGPD in terms of raw route-crunching throughput.

Cheers,

Marko



More information about the Xorp-users mailing list