> > So, with 64MB or so, it's possbile to store fairly complete routing
> > tables for IPv4. The problem is the larger address blocks of IPv6.
> > (And also that some older routers can't use so many simms. This is not
> > really a problem for most Linux systems tho.)
>
> The killer isnt just routing tables. The killer is lookup time. Cisco stuff
> is supposed to flake out at about 40,000 routes from figures people have
> quoted. Hence the fact sprint and folks won't route smaller than a /19 and
> keep making noises abou going to /17.
It the lookup time a problem because the physical memory is slow, or is it a
lack of processor power? It seems to me that throwing money at the problem
would fix either one. Using 15ns cache ram (sram?) instead of the slow
stuff would be quite a boost (if an expensive one) to the memory access
time. Unless they already use fast memory. I'd be interested in hearing
about the hardware used in dedicated routers.
And wouldn't a 16 processor board filled with 200MHz P6s be sufficient for
processing power? Just how much horsepower does your typical router have
anyway?
And, yes, i realize i'm probably not the first person to think of these, but
i don't understand the problem very well, so perhaps someone could fill me
in a bit, or direct me to where to look. Not to mention a better forum than
this (comp.hardware.routers anyone?).
> I would imagine some other kit
What do you mean by "kit" here?
> can do better some worse, nevertheless trying
> to look up destinations from 40,000 choices at 100Mbits/second on 10 network
> ports, while recomputing the routing tables is NOT a trivial problem.
-- nathan wagner nw@hydaspes.if.org