[MLB-WIRELESS] anyone alive out there?

Richard James Salts melwireless at spectralmud.org
Thu Sep 5 11:40:25 EST 2013


On Thu, 5 Sep 2013 11:19:36 Mike Everest wrote:
> > I think I'd do it differently than both proposals. Active network with
> 
> ipv6 ULA to
> 
> > the end user. VPN for retail isp providing transit to people out of the
> 
> network.
> 
> > We can do better than a layer 2 protocol.
> 
> Interesting concept, but I don't understand why a layer 3 wholesale carriage
> network would be /better/ than a layer2 network...?
> 
> Selling layer2 service allows the retailer (and potentially the customer) to
> lay any scheme they want over the top.  Even if ia customer wants a layer
> 22 service (for example between two branch offices) then it can be done
> with the current delivery model.
> 
> If the wholesale network was layer3 (ipv6 OR ipv4) then if you want a layer2
> service, you'll need to lay a tunnel on top of the IP - i.e. burn up some
> more overheads...
> 
> As it stands, IPv6 ULA to the end user can easily be delivered by a retail
> ISP who (of course) controls and manages their IPv6 space.
> 
> And the killer: If the wholesaler delivers the IP network, then how would
> the address space be managed so that it would be routable by any arbitrary
> ISP?  It would make a mess of the routing efficiencies (supposedly)
> delivered by IPv6 scheme...

I was just basing it off the research into datacentre networks that's out 
there. There are a few papers about Jellyfish networking for redundancy and 
throughput. I think http://www.cs.illinois.edu/~pbg/papers/jellyfish-nsdi12.pdf 
was the original paper proposing the network topology. Running a large layer 2 
network also means that you either go with a point to point architecture or 
have a large broadcast domain. There is also the fact that getting routing 
redunancy is much more easily achieved in a layer 3 network with protocols 
such as OSPF and BGP well understood and widely deployed. Equivalent solutions 
in ethernet such as spanning tree are often a lot slower to converge in the 
event of failure.


> 
> Fair thought experiment though! ;-)
> 
> Cheers!


More information about the Melbwireless mailing list