From nordunet.redist.ietf Thu Aug 12 11:00:52 1993 Path: news.daimi.aau.dk!uts!sunic!sunet-gateway!owner From: ses@tipper.oit.unc.EDU (Simon E Spero) Newsgroups: nordunet.redist.ietf Subject: Re: address sizes , and the radical side IPNG Date: 12 Aug 1993 05:23:17 +0200 Organization: Royal Institute of Technology (KTH), Sweden Lines: 68 Sender: daemon@sunic.sunet.se Message-ID: <9308120307.AA20476@tipper> References: <9308120033.AA09964@aland.bbn.com> Craig Partridge writes: > > I'm happy to cope with small variable addresses, or (if I must) really >big fixed sized addresses. I'd really like to avoid big and variable. >(Note this is from the perspective of someone who is spending most of his >time doing gigabit networking research and thus worries about things like >how to forward hundreds of millions of IPng packets per second). > If we must have a fixed 64bit quantity, why not let it encode the MTU size :-) Actually, I think it's an EID thing. We need something relatively short so we can easily identify the destination for a packet. That's what EIDs are for. We need something to tell us where that endpoint lives - addresses . Then we need something to tell us how to get there -routing doo-dabs. I find it much easier to envisage a scalable system based on 'flow-a-grams'- mostly switching on EIDS and ignoring the address unless it changes. In a multimedia world, the bulk of traffic will be acting in a connection oriented way, with resource reservation, and a radically different approach to packet loss and transmission errors - if it's not there in time for the next frame, forget it, I don't want it - and don't expect a tip. In such a world, the ability to have variable length addresses which can be created dynamically at setup, and change during the life of the session seems essential. Resource reservation could have a massive effect on routing paradigms. In the current paradigm, there seems to be an implicit assumption that if a packet to a certain destination took a particular route, then that is a good route to take for another packet. With resource reservation, packets to the same destination but on different 'flows' are likely to take radically different routes. Suppose I borrow a few million dollars, and run some fibre along the railway tracks from Raleigh to Washington, and stick some switches and routers onto each end. Now, since any bandwidth I don't sell is gone for good, I'll want to advertise at a good price to sell. However, since I'm taking reservations (higher price, more profit), I don't want to over commit. Any policy based routing system worthy of the name will have to support some sort of true Routing Information Exchange, as well as a Routing Information Futures Market - "One point five megabits RDU-WAS, 10:15 - 10-30; offer 256, bid 254". Imagine if 50 other people had the same idea. Imagine none of them having enough guaranteed bandwidth to meet a request, so a route via charlotte is needed. For a given connection, the path might remain fairly constant, modulo the occasional outage, or a blue-light special on another link, but across connections there'd be huge differences. Message: address is just a source route with nothing left to loose. Radical IPNG- [ On behalf of the children of the Thatcher-Reagan era, I'd like to make the following statement; you spent our money for your tax cuts, and take the payments for your pensions. You built and bought your cars with dirty engines, and took them to the drive-thru so we could serve you burgers. You cut down all our forests, and what you couldn't cut, you poisoned. Now it's time to make another choice, and you say "Not in my lifetime". Not in your lifetime, In our lifetime. Make the right choice. The power to choose. ] ----- Hackers Local 42- National Union of Computer Operatives, Chapel Hill section ------------------------------------------------------------------------------ Tar Heel Information Services - Nothing but net! | WAIS/Z39.50 spoken here CLNP - The C is for Clue | DoD #612 | Tel: +1-919-962-9107