[quagga-dev 13344] Re: VRF device integration in quagga
Michael H Lambert
lambert at psc.edu
Tue Oct 20 15:24:54 BST 2015
> On 20 Oct 2015, at 08:09, Paul Jakma <paul at jakma.org> wrote:
> Just as an aside, there is no known way to achieve:
> Layered routing AND Scaling AND Efficient forwarding
> for all hosts on arbitrary networks (e.g. the Internet). It'd be good if there more awareness of this in the routing engineering world. :)
> There are many sound reasons to use split tables and layering of routing, but generally scalable routing just isn't known to be one of them (least, not while retaining efficient routing).
I'm not sure this is "just an aside". It think it's the fundamental point. There is a minimum level of complexity in any system and routing is a compelling example of that. That complexity can often be moved to different parts of the system, but it's always present. That's why I'm not convinced that it's any better (or worse) to have virtualization within quagga on a single machine versus running separate instances of quagga on multiple machines (physical or virtual). When dealing with hardware-based routers costing tens or hundreds of thousands of dollars it's an easy call, but when running on commodity hardware it's not so clear cut.
More information about the Quagga-dev