-----Original Message----- From: Owen DeLong Sent: Saturday, April 03, 2010 9:13 PM To: Zaid Ali Cc: nanog@nanog.org Subject: Re: legacy /8
On Apr 3, 2010, at 2:49 PM, Zaid Ali wrote:
They are not glowing because applications are simply not moving to IPv6. Google has two popular applications on IPv6, Netflix is on it way there but what are other application companies doing about it? A popular application like e-mail is so far behind [ref: http://eng.genius.com/blog/2009/09/14/email-on-ipv6/] and I still encounter registrar's providing DNS service not supporting Quad A's.
Uh, netflix seems fully functional to me on IPv6. What do you think is missing?
Well, there are a lot of companies out there building server applications who don't have the human bandwidth to migrate the application. Things like application clusters that do a lot if internal multicast, for example, would need to change. My instincts say you are going to see v6 roll out in a lot of production networks with devices like load balancers that can have an IPv6 virtual IP balanced to v4 servers on the back end. In that sense people aren't really migrating to v6 so much as they are accommodating it. And the vendors have some real problems when it comes to v6 that derives from lack of resource allocation. I can put together a bodacious route server on a quad CPU (hex core) with almost 200Gig of RAM for less than the cost of a management module on a small-ish router and offload a lot of the BGP stuff from the routers. In fact, with a few minor tricks you can have your edge routers connected to several peers and not have the peers talking BGP to your routers at all, they are all talking to your "bodacious" route server to which you can add processing power / RAM as needed. At that point your edge routers have one single BGP peer, the route server. And I have actually used this sort of thing in production albeit "back in the day" when the routing table first started to explode beyond the capabilities of some purple cyclops layer 3 switches I was using at the time. That is the fundamental problem as I see it right now. The vendors do not want to ship the resources required until they need to (it isn't a problem until it is a problem) and they all end up behind the curve reacting to this crisis or that rather than getting ahead of the game and producing the router than can handle a serious number of 128-bit routes. It isn't that the routing policy is bad, it is that we continue to have to work around the limitations of the gear shipped from the vendors.