>I suppose this is a bug in linux...but I can't traceroute to x.x.x.0.
>Is this some left over breakage from classful IP routing?
Maybe, maybe not. True classless routing means you can use the first
subnet and the last subnet, that you can break subnets on virtually
any bit boundary. But I don't think that you should assign the network
address (such as .0) to a host if you can possibly avoid it. And if you
are talking about a broadcast medium like ethernet, you also have to
reserve the last (broadcast) address.
>Even from my Ciscos, I can't do a trace ip x.x.x.0.
So maybe its not just linux.
>To me, it seems the route calculator in the PM is broken. A starting
>address of 205.229.60.1 and pool size of 30 is 205.229.60.0/27, leaving
>the first IP for network address and last address for broadcast. When I
>setup gated on my Linux term servers, I will be aggregating in such a way.
Its not broken, its just "different" from common usage. When you set your
starting address to .1, thats where the Livingston starts to calculate its
best fit routes. NOT at .0 And it doesn't even try to make a neat /27.
So you end up with 5 or so subnets instead of a neat one. Livingston uses
its "private" knowledge that we are referring to a non-broadcast medium to
do away with the network address and broadcast address. You don't need a
broadcast address as this is not a broadcast medium but point to point.
And for some apps you don't strictly need a network address. Like I said
earlier, I would like the option to say that if I give an assigned address
of 205.229.60.1, that the network address is one-less than the assigned
address. So that in the most common case of the PM-2, you get neat /27's.
For the PM-3's you can do a /26, but unless you are loaded with 60 modems
you waste several addresses. Frankly if you use the PM's for telnet
and/or rlogin as well as PPP/SLIP, you are wasting addresses anyway. You
could set the pool to the actual number of modems so that higher addresses
aren't used, and then use your "wasted" addresses for use as static IP's
with /32 netmasks. You would have to have these individual host routes
anyway. I can see the use of the best-fit, to avoid wasting addresses.
But several network products out there can't use/understand the use of the
.0 address as a single host. This is analogous to the use of the zero
subnet. Maybe some day, it will have common usage, but for now, it is
just bad manners.