<br><br><div class="gmail_quote">On Sat, Aug 20, 2011 at 2:43 PM, Lukasz Stelmach <span dir="ltr"><<a href="mailto:stlman@poczta.fm" target="_blank">stlman@poczta.fm</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
W dniu 20.08.2011 19:37, Sergei Zhirikov pisze:<br><br>
In the simplest setting (one uplink, one router, one LAN)<br>
radvd+hosts act exactly the same way as routed broadcasting RIP<br>
messages to hosts running RIP client listening to them (nb. Win98 or<br>
even 95 had a RIP client). Routing daemons and protocols provide<br>
enough information to build a complex routing table but for a LAN<br>
with a single gateway information radvd provides is just enough.<br><br></blockquote><div><br></div><div>I wonder if this is a matter of framing the responsibilities of each component</div><div>to illuminate the assumptions in contemporary network routing. If I may.</div>
<div><br></div><div>In most networks, hosts only know a couple of things. They are assigned</div><div>or they discover their IP address. The have a default gateway. They have</div><div>access to a name server. And in many cases they optimize local traffic</div>
<div>due to the fact that they know the network address of their directly connected</div><div>ethernet segment. This isn't mandatory, but it is typical.</div><div><br></div><div>Routers act as gateways for hosts and for other routers. They know the</div>
<div>network prefixes that are accessible on each interface. They also have</div><div>a default gateway, a route of last resort, which is the uplink</div><div>for leaf networks. In a multiple uplink setup there could be several</div>
<div>gateway routes though one may be used exclusively as is</div><div>probably the case with an IPv6 tunnel.</div><div><br></div><div>Radvd is concerned with the self-configuration of host IP addresses, with</div>
<div>selection of the default gateway for hosts, and discovery of the name server.</div>
<div>DHCP, too, delivers this information to hosts via a different kind</div><div>of mechanism.</div><div><br></div><div>It is a coincidence, that radvd appears to be able to advertise inter-network routes</div><div>as one might expect a router to do. Routers will probably take on the</div>
<div>duty of advertising IPv6 prefixes just as routers serve DHCP. But that is not</div><div>an avenue for blending routing with IP address configuration/assignment.</div><div><br></div><div>So, here's the crux. Switches and routers have merged such that companies</div>
<div>commonly deploy a backbone switch with routing capabilities. In these networks,</div><div>there is no advantage for hosts to know anything about routes. The switch is the</div><div>gateway and it has a fast electronics for switching and routing packets.</div>
<div><br></div>
<div>Home networks tend to use less capable equipment. There may be an access point</div><div>with some routing capabilities, but it is unlikely to be as capable as a backbone switch.</div><div>It probably doesn't terminate the IPv6 tunnel and it probably cannot perform route</div>
<div>advertisement vis a vis radvd. Thus, the IPv4 gateway would be different from the</div><div>IPv6 gateway.</div><div><br>
</div><div>It's tempting to think that we could optimize routing in a small network</div><div>such that each host knows the best route for traffic. But there may be no real</div><div>advantage to this. Consider that uplinks are probably less than 10Mib and switches</div>
<div>and hosts have 100Mib interfaces. An extra hop will probably introduce a small</div><div>latency in the traffic, but the router won't be a throughput bottleneck. </div><div> </div><div>So, the apparent simplicity of the corporate network isn't going to be available to</div>
<div>home networks for some time. I've resorted to building my routing tables</div><div>by hand. Fragile? Somewhat. Effective? Sure.</div><div><br></div><div>Cheers</div></div>