query-source for multiple interfaces

Sam Wilson Sam.Wilson at ed.ac.uk
Wed May 24 13:46:53 UTC 2006


In article <e4lmku$18e9$1 at sf1.isc.org>,
 Barry Margolin <barmar at alum.mit.edu> wrote:

> In article <e4km6t$2q0b$1 at sf1.isc.org>,
>  Sam Wilson <Sam.Wilson at ed.ac.uk> wrote:
> 
> > In article <e4j9g9$1i2i$1 at sf1.isc.org>,
> >  Barry Margolin <barmar at alum.mit.edu> wrote:
> > > Also, unless the server is running a routing daemon, it probably won't 
> > > take much advantage of the fact that it's multihomed.  It will just send 
> > > most of its queries through its default gateway, so you might as well 
> > > set query-source to the interface connected to that.
> > 
> > Of course it's running a routing daemon - how else would the anycast 
> > address be propagated?
> 
> From a static route on the upstream router, as described in one of the 
> slides in the presentation you linked to earlier.  That's how we did it 
> when I was at Genuity.

Our reason for doing it is to provide a more reliable local service.  At 
present if a server goes down then any dumb resolver with one of that 
server's addresses in its resolv.conf or equivalent will have to timeout 
on each query.  Going for anycast with automatic withdrawal of the 
anycast prefix on server failure (or routing failure) should ensure that 
a server will always respond.  Static routing wouldn't have that effect, 
a situation also mentioned in the slides.

(I guess you know all that - I'm really filling in for any other 
interested parties who may be lurking.)

Sam Wilson                             one of hostmaster at ed.ac.uk
Infrastructure Services Division
Computing Services, The University of Edinburgh
Edinburgh, Scotland, UK



More information about the bind-users mailing list