> For simple curiosity, I'd like to know why you don't think it qualifies as a
> true gigabit router.  The network layer (if I remember my OSI model
> correctly) is the IP level, correct?  How can the router not be aware of the
> IP layer?  I probably don't understand what you mean by the term aware.
>> Gabe

Well let's put it this way. I won't say anything for the purpose of
being dictionary correct, but I've been involved in IETF work for a
few years now and even have some credit on an internet standard, and
here's what I would say.

A switch, which not really more aware, typically caches ARP (Address
resolution protocol) requests to associate a particular IP that is
immedeatly connected to it with a particular phsyical (MAC) address,
so if it get's data for 192.168.1.15 and it knows that's associated
with MAC blah on port 4 then it will only send that traffic there.
This in no way implies that the switches knowledge and cacheing of ARP
allows it to deal with subnets or anything like that.

A ROUTER is able to take traffic not destined for any immedeatly
connected nodes and make an apporpriate "routing" decision. So lets
say you have 2 routers, and edge router connected to your WAN/ISP and
another internal router that deliniates a network for your children
with the same IP address space (not NAT'd) as the rest of your
internal network. If a packet comes in for one of your "child" PCs, if
that box was a multi port ROUTER then it would not send the data
anywehere but the appropriate connection which is in fact the other
router at the edge of your "child" network, whereas in this case, at
least on the first try the switch portion of that router will be
transmitting all of the packets on every port on your network. Granted
once the ARP cache associates an IP/MAC with a certain port this may
stop, but in fact if all of the ports on the device were "routed" then
that would never happen the first time. The exchange of BGP/OSPF
information would tell the first router, that certain IPs or a certain
IP address range lie one hop out on this certain link via this certain
IP. Maybe an even simpler example is the default path, where your
broadband router takes anything not destined for the internal network,
and forwards it to your WAN link. Most home routers in fact do just
this, and don't even really transmit/receive information about the
network around them by default I think.

So in summary, for a device to be netowrk "aware" it needs to have a
method for exchanging information about the netowork layout, often via
BGP(border gateway protocol) or OSPF(Open Shortest Path First) routing
algorithms, and a method to act on such information by routing packets
to the appropriate interface and nowhere else, usually trying to
optimize heuristically the best path, because as we know the
travelling salesman problem is NP complete, and this is just a
sub-problem of that, so still NP complete. A switch on the other hand
simply is able to track the traffic that has gone through it and
occasionally contrain later traffic to certain addresses to certain
ports. A hub (or DSL bridge for example which just bridges physical
media) doesn't even do that, and simply blindly retransmits data
everywhere. So clearly there's a lot more out of band communcation
inolved with routed ports, more intelligence in the decisions before
forwarding (why a wire speed switch is simple and cheap but a wire
speed gigabit router is VERY expensive), and frequenetly for example
at gigabit speeds will involve special types of memory in the router,
and lots of optimized network hardware.

Hope that helps. I'll put 95% confidence level on the minor details.

-Tom


--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to