On Sat, 14 Mar 1998, Alec H. Peterson wrote:
At 02:36 3/14/98 , Marc Slemko wrote:
For example? A router with one ATM interface going to the world with a high MTU with an ethernet on the other side. Say you use private IP space for links on that router. Say someone on the Internet filters traffic from private netblocks; lots of people do. There _can_ be machines that are completely unable to transfer data (eg. download a web page) from another because you just broken path MTU discovery. This is not a made up situation, this is a real example that I have had to deal with of how using private IP space for network interfaces used for public traffic does break things in some situations.
You only run into this situation if:
1) The packets have DONT_FRAG flag set on them
As I said, path MTU discovery. That imples DF.
2) The ATM interface is in fact set with a small MTU.
1 is very possible nowadays, but 2 is prolly not. Yeah, ATM has a small cell size, but most IP over ATM interfaces I've ever seen have an MTU of something like 4470. Yes, they cheat and do break down the packet into cells, but can you imagine trying to put a TCP download into 48 bytes? You'd use most if not all of that for the IP header.
No, the whole point is the ATM interface has a large MTU and the ethernet has a small MTU, which means that large segments coming from the "outside" to the "inside" don't fit. This is just a simple example from my life; there are many situations that can cause this which aren't always obvious.
It is still an open debate about whether or not RFC1918 space is wise to use, but I'd say it's a sign of a commendable effort on @Home's part that they are trying to conserve IP space, even though they do have lots of routable addresses.
You can call it an open debate until you actually try using it. Of course, most people don't notice the things that break when they do.