Re: IPv6 adoption
On Mon, Jul 17, 2000 at 12:21:18PM -0600, Jason Gunthorpe wrote:
> Nope, the really high speed routers use switching, once a route is
> computed by the CPU the gate arrays store that route and just switch
> packets.
>
> This is sometimes called layer 3 switching IIRC.
Have to be careful with those marketing terms ;)
OK what usually happens with routers is they look up the route using the
patricia tree, which is like a boolean tree but has variable bit width
so is slower (you have to do some ANDing as well), check the Qos, check
this that, ACLs etc then forward the packet.
The idea is that some routers hash all those fields together and have
some sort of hash table, match the key and you get forwarded right away.
It's basically how NetFlow works (ok simplified but you get the idea).
If all routes were the same length and they weren't too big, you can do
neat stuff like use CAM (Content Addressable Memory) which is a bit like
cache memory only you don't use memory location as the key but
destination address of routes. Probably routes are too huge today for
this, but this is why you get a performance increase with tag/MPLS
though it is only a little increase.
I'd say the IETF is looking at these first 64 bits and thinking
something useful like MPLS could be done to it. In any case having a
fixed length route would be interesting.
- Craig
[Off topic comment: And like when is Debian getting a central IPv6 repository?]
--
Craig Small VK2XLZ GnuPG:1C1B D893 1418 2AF4 45EE 95CB C76C E5AC 12CA DFA5
Eye-Net Consulting http://www.eye-net.com.au/ <csmall@eye-net.com.au>
MIEEE <csmall@ieee.org> Debian developer <csmall@debian.org>
Reply to: