RE: Sequentially assigned IP addresses--why not?
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Anthony, Let me try and say this kindly (since after it is pointed out several hundred times it gets quite frustrating). If you don't see the processing requirements then you have *no* understanding of how routing works. Forwarding packets is easy. Routers basically have two related but completely seperate functions. The *second* function is to take an incoming packet, look at its destination address, compare it to its forwarding table, and forward the packet. This is easy in the whole scheme of things. The *first* function is to calculate that forwarding table used above. This is the place that processing power is needed. On busy routers this calculation can be necessary hundreds of times a second. The more routes the larger the process of recalculating the forwarding table when a change *anywhere* in the topology occurs. - --- Phil P.S. Go find the CIDRD archives, or (if you want even earlier) try the big-ip list to see the debate dozens of times. -Original Message- From: Anthony Atkielski [mailto:[EMAIL PROTECTED]] Sent: Friday, August 11, 2000 9:38 PM To: [EMAIL PROTECTED] Subject: Re: Sequentially assigned IP addresses--why not? We seem to be talking 5-6 orders of magnitude in speed here. Even Moore's Law doesn't help in that range. I don't see why all this processing power is required. You look at the incoming address, you figure out which outbound path can handle that address, and you forward it. Simple. Even if the full address is a thousand digits long, you only have to look at the digits around your level to determine the next step in routing. -BEGIN PGP SIGNATURE- Version: PGP 6.5.3 iQA/AwUBOZT4YR8Cp2AdP9rUEQKQIwCfQUa6eJOLwwXGgXEgcxw4cMHQ/NYAnAwx ZLxXZHam4Ns5cxWJy4zaA5sB =Rdyp -END PGP SIGNATURE-
Re: Sequentially assigned IP addresses--why not?
Let me try and say this kindly (since after it is pointed out several hundred times it gets quite frustrating). If you don't see the processing requirements then you have *no* understanding of how routing works. You need not go to great pains to be "kind" about it. This is a pretty standard preamble to a post that essentially means "I disagree." The *first* function is to calculate that forwarding table used above. This is the place that processing power is needed. On busy routers this calculation can be necessary hundreds of times a second. The more routes the larger the process of recalculating the forwarding table when a change *anywhere* in the topology occurs. I wonder how human beings manage to route their cars from one point to another, given how much more slowly they process things than do routers.
Re: Sequentially assigned IP addresses--why not?
"Steven M. Bellovin" wrote: In message 003b01c003c6$3ffe9230$0a0a@contactdish, "Anthony Atkielski" wr ites: The telephone company has milliseconds to seconds to resolve an address into a route. The Internet has microseconds to nanoseconds to do so. Build faster hardware. We seem to be talking 5-6 orders of magnitude in speed here. Even Moore's Law doesn't help in that range. Also, circuit setup needed for establishing routing labels requires at least one round trip time (ignoring processing time), which is pretty much a constant given geographic distance. -- Henning Schulzrinne http://www.cs.columbia.edu/~hgs
Re: Sequentially assigned IP addresses--why not?
This is a stupid argument. An IP packet (for instance) relies on external decision-making (the router) while a car has an intelligent (or at least decision-making) human inside, making the "car+human" unit self-supporting in terms of route decision. So?
RE: Sequentially assigned IP addresses--why not?
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 -Original Message- From: Anthony Atkielski [mailto:[EMAIL PROTECTED]] Sent: Saturday, August 12, 2000 3:44 AM To: [EMAIL PROTECTED] Subject: Re: Sequentially assigned IP addresses--why not? I wonder how human beings manage to route their cars from one point to another, given how much more slowly they process things than do routers. Trying to dismiss a technical problem with a facile analogy that is not relevant is not a way to pursuade. Roadways to not change topology dozens of times per second. Cars do not travel at a large percentage of the speed of light so calculating the route to your corner store usually doesn't involve considering routes through Denver, New York, Dallas, Chicago, San Jose, Washington DC and Atlanta all at the same time. People don't often mathematically optimize their routes based on traffic flow patterns that are being inputed from hundreds of sources per second. I could go on, but what is the point? If you want to participate on a mailing list devoted to engineering of Internet protocols please don't expect to be listened to seriously unless you are prepared to conduct the discussion on a technical level. - --- Phil -BEGIN PGP SIGNATURE- Version: PGP 6.5.3 iQA/AwUBOZXc9B8Cp2AdP9rUEQILmACg7CvmV8MmGyTyO/fD88KyWvMe69YAn1y0 L7zIuHr8nCQV3a11LyTZBIGa =HnjF -END PGP SIGNATURE-
Re: Sequentially assigned IP addresses--why not?
Trying to dismiss a technical problem with a facile analogy that is not relevant is not a way to pursuade. Logically, then, calling the analogy "stupid," or saying "we looked at it and decided it wouldn't work," is even less so. In any case, I have no objective to persuade. I'm sure that all the decisions have been made. I'm confident that the passage of time will provide all the persuasion needed. Roadways to not change topology dozens of times per second. It depends on the time scale of observation. I could go on, but what is the point? I agree. If you want to participate on a mailing list devoted to engineering of Internet protocols please don't expect to be listened to seriously unless you are prepared to conduct the discussion on a technical level. There are certainly some people who probably should not listen to me, as it will just raise their blood pressure. However, there are also others who are willing to listen to anyone, just as I am. Those in the former category can censor at their receiving end, so there is no reason to censor at the sending end, particularly since it would be to the detriment of people in the latter category. Speaking of technical levels, what level does it require to change the Reply-To address on a mailing list so that replies go to the group by default, instead of to the sender of whatever message elicits a reply?
Re: Sequentially assigned IP addresses--why not?
and drive right into the bay over that bridge that your GPS insists exists. My GPS has no knowledge of bridges. I look out the window to see if there are any obstacles. I just follow drivable paths in the general direction of the heading shown on my GPS until I arrive. It is surprisingly efficient, and I don't need any map at all. Where do you think your GPS gets its info? From satellites overhead.
Re: Sequentially assigned IP addresses--why not?
On Fri, 11 Aug 2000, Anthony Atkielski wrote: The problem is that we (as a profession) don't know how to do that. We have to make routing scale, and that demands aggregation, which in turn demands structured addresses. The telephone company figured out how to avoid problems decades ago. Why the computer industry has to rediscover things the hard way mystifies me. Oh god, not this argument again. This is the circuit vs connectionless debate. I am sure if you do a search on Kleinrock and Mills in open literature, you will find all sorts of reasonings behind why this divide exists. To grossly oversimplify things, the phone systems do a relatively slow setup and once it is set up, let it stay till it is done and then tear it down. There isn't a phone company that does setups and teardowns (if I may stretch the term) at a rate that can match the connections initiated and torn down involving tcp/ip for http alone that pass through a core router in any promising local ISP. See ATM to the desktop. /vijay
Re: Sequentially assigned IP addresses--why not?
Anthony Atkielski wrote: Brian Carpenter writes: This is some sort of urban legend. If a routeable prefix was given to every human, using a predicted world population of 11 billion, we would consume about 0.004% of the total IPv6 address space. Surely you recall the quotation attributed to Thomas J. Watson: "The world will never need more than five computers." Indeed, although he probably never said it. That's why we didn't pick 64 bits for the IPv6 address. Brian
Re: Sequentially assigned IP addresses--why not?
"Rakers, Jason" wrote: This is some sort of urban legend. If a routeable prefix was given to every human, using a predicted world population of 11 billion, we would consume about 0.004% of the total IPv6 address space. that's what they said about never needing more than 640kb of memory in a computer.. we'll never need more than that! Please think very carefully about the orders of magnitude involved here. The two cases are not comparable.
Re: Sequentially assigned IP addresses--why not?
"Corzine, Gordie" wrote: Look, my days as an engineer are a distant memory, so I won't try to work this out in detail. Maybe there are irrefutable reasons why this can't be done, but I do believe the current architecture will lead to premature exhaustion of the address space. It will take far longer to design and deploy something that is so technically elegant it solves all problems and pleases everyone. At some point you simply have to move forward. To do nothing can be far more dangerous (as proven by the disdain for NAT). Can IPv6 be worse for the net than NAT? If premature depletion of IPv6 addresses is the biggest problem IPv6 ends up encountering I'd say the net is in good shape. It's probably more likely that new problems no one had considered will arise. I see rough consensus, move forward. John
Re: Sequentially assigned IP addresses--why not?
Sean, I agree with you. I was trying to make it simple. Brian Sean Doran wrote: Brian Carpenter writes to Anthony Atkielski: | The telephone company figured out how to avoid problems decades ago. Why | the computer industry has to rediscover things the hard way mystifies me. | | The telephone company has milliseconds to seconds to resolve an address | into a route. The Internet has microseconds to nanoseconds to do so. You are missing the difference between "what" and "where". The telephone company takes milliseconds to translate the equivalent of 6.6.9.9.9.6.6.8.6.4.e164.net into the equivalent of 192.36.143.3. That is, the phone number is merely an identity name, which is converted into a location name by a database lookup. The principal difference between hop-by-hop packet-based networks and circuit-based networks is that in the former the location name does not require negotiations among the intermediate systems, or between the first-hop IS and the originating end system. There is a simple assumption that each hop will be able to make a reasonable forwarding decision on any location address, even if the location address is "unexpected". In circuit-based networks, this is not generally the case. The means and costs of translating a "what" address into a "where" address are often strikingly similar in both circuit-based and hop-by-hop packet-based networks. Sean.
Re: Sequentially assigned IP addresses--why not?
At 11:43 PM -0400 8/10/00, Vijay Gill wrote: On Fri, 11 Aug 2000, Anthony Atkielski wrote: The problem is that we (as a profession) don't know how to do that. We have to make routing scale, and that demands aggregation, which in turn demands structured addresses. The telephone company figured out how to avoid problems decades ago. Why the computer industry has to rediscover things the hard way mystifies me. Oh god, not this argument again. This is the circuit vs connectionless debate. I am sure if you do a search on Kleinrock and Mills in open literature, you will find all sorts of reasonings behind why this divide exists. To grossly oversimplify things, the phone systems do a relatively slow setup and once it is set up, let it stay till it is done and then tear it down. There isn't a phone company that does setups and teardowns (if I may stretch the term) at a rate that can match the connections initiated and torn down involving tcp/ip for http alone that pass through a core router in any promising local ISP. Actually it has less to do with the connectionless/connection debate and more to do with what they are naming. What the phone companies did (and the Internet as yet to do), was precisely what John Shoch outlined in his paper over 20 years ago and that Saltzer expanded on not quite 10 years later. They made the location independent "addresses" application names and kept the location dependent names, i.e the addresses on which they do their routing. These network addresses are the same as the ones they have always been using but they are only internal. It is just that they used a similar syntax for both to give people the impression they were actually doing something else. But the syntax of the names has nothing to do with their semantics. Now, it is the case that most communication with applications both in the phone system and on the Internet is connection based so this mapping does not have to be done too often. So there is a connection (no pun intended) but it is distinctly secondary. However, it remains that it is applications that should have location independent names and network addresses that should have location dependent names. Take care, John
Re: Sequentially assigned IP addresses--why not?
Sean; Brian Carpenter writes to Anthony Atkielski: | The telephone company figured out how to avoid problems decades ago. Why | the computer industry has to rediscover things the hard way mystifies me. | | The telephone company has milliseconds to seconds to resolve an address | into a route. The Internet has microseconds to nanoseconds to do so. You are missing the difference between "what" and "where". The telephone company takes milliseconds to translate the equivalent of 6.6.9.9.9.6.6.8.6.4.e164.net into the equivalent of 192.36.143.3. That is, the phone number is merely an identity name, which is converted into a location name by a database lookup. In that sense, DNS names are randomly (more aggressive than sequentially) assigned addresses. Masataka Ohta
Re: Sequentially assigned IP addresses--why not?
John Kristoff [EMAIL PROTECTED] writes: | To do nothing can be far more dangerous (as proven by the disdain for NAT). The disdain for NAT is non-uniform. Personally, I rather like NAT. | Can IPv6 be worse for the net than NAT? IPv6 and IPv4 will coexist for a time; the topology of the (large) IPv4 Internet and the (tiny) IPv6 Internet are discontiguous, and is unlikely to cease being so before IPv6 curls up and dies. There are real operational costs to maintaining ships-in-the-night multiprotocol networks; the maintenance cost of such networks is one factor in why we don't see DECNET Phase IV, IPX or CLNS being forwarded by equipment in the core of the IPv4 Internet. NAT and inter-protocol header translators (e.g. FAITH or 6to4, ironically written by Carpenter and Moore, who both really hate NAT) totally eliminate the near-term need to even consider ships-in-the-night in the core. They also can reduce the weak pressure on the IPv4 address space by aggregating multiple hosts behind a single (IPv4) address. Sean.
Re: Sequentially assigned IP addresses--why not?
Fred Baker asks: | When I build a telephone out of an IP dialler attached to | someone's waist, a modulator on their necklace, and an earphone attached to | their earring, all connected by IP on BlueTooth, what addresses do I put on | the different components of the telephone? RFC-1918 for all but one "outside" address on the component which does NAT. Note that the "outside" address may also be an RFC-1918 address, and ideally should be gotten automatically via DHCP. Better question: what are the DNS names of the components, and how are they published to the "outside" world? Sean.
Re: Sequentially assigned IP addresses--why not?
The telephone company has milliseconds to seconds to resolve an address into a route. The Internet has microseconds to nanoseconds to do so. Build faster hardware.
Re: Sequentially assigned IP addresses--why not?
- Original Message - From: "Brian E Carpenter" [EMAIL PROTECTED] To: "Corzine, Gordie" [EMAIL PROTECTED] Cc: [EMAIL PROTECTED] Sent: Thursday, August 10, 2000 9:30 PM Subject: Re: Sequentially assigned IP addresses--why not? "Corzine, Gordie" wrote: Seriously, As was pointed out recently, IPV6 will croak much sooner than it needs to for the simple reason that we structure routing intelligence into the address assignment. This is some sort of urban legend. If a routeable prefix was given to every human, using a predicted world population of 11 billion, we would consume about 0.004% of the total IPv6 address space. (The actual calculation is 11*10^9/2^48 since there are 48 bits in an IPv6 routing prefix. Or 11,000,000,000 / 281,474,976,710,656 = 0.39 ) Does this mean that every router will have to handle 2^48 routing table entries and that this vast amount of information must be sent over the internet on every routing table update? Salavat
Re: Sequentially assigned IP addresses--why not?
Hello What is the difference between plain address (I mean house address like 47 Ulcombe gardens, Canterbury, Kent, United Kingdom) and IP address. The former is scalable to whatever size one may want and the router for plain address (i.e. post office in USA, for example) does not have to know about 47, Ulcombe and so on it only must know what direction UK is located. Why not to take this analogy and use it in the Internet. The difference is not very big, since plain mail system is connectionless. Regards Salavat
Re: Sequentially assigned IP addresses--why not?
Phone numbers have moved from being direct as originally implemented to being a level of indirection, thanks to a lot of behind-the-scenes mucking about. The Internet introduced DNS to gain that same level of indirection. Phone numbers are now portable; DNS names are portable. I don't agree with that. Host names, and a means for translating them to addresses, existed before DNS. Introduction of hierarchical naming and DNS let the maintenance of this translation mechanism be decentralized. Hm, wasn't this thread started by a suggestion that so-called addresses be assigned under centralized control?
Re: Sequentially assigned IP addresses--why not?
At 04:40 PM 8/10/00 -0400, Steven M. Bellovin wrote: Look at it this way. We have about 75K routes in the "default-free zone" now. No - that was March 2000 - now we have about 87,000 (www.telstra.net/ops/bgp) If we just assigned addresses sequentially, we'd need a route for every endpoint. There are what, 100,000,000 nodes today, and more tomorrow? We can't handle 3 orders of magnitude increase in the size of that table, let alone what it will be in a few years. There are a number of scenarios which will make the routing system crash and burn - this is one of them. On the other hand even doing nothing will be a problem - we appear to have resumed exponential growth of the routing system again, presumably as multi-homing at the edges starts to be more and more common. Geoff Huston
Re: Sequentially assigned IP addresses--why not?
Try reading one of the books on Internet routing, there are several good ones. Brian
Re: Sequentially assigned IP addresses--why not?
Does this mean that every router will have to handle 2^48 routing table entries and that this vast amount of information must be sent over the internet on every routing table update? Salavat In a word, no. In two words, Hell no! See RFC 2374.
Re: Sequentially assigned IP addresses--why not?
In message 003b01c003c6$3ffe9230$0a0a@contactdish, "Anthony Atkielski" wr ites: The telephone company has milliseconds to seconds to resolve an address into a route. The Internet has microseconds to nanoseconds to do so. Build faster hardware. We seem to be talking 5-6 orders of magnitude in speed here. Even Moore's Law doesn't help in that range. --Steve Bellovin
Re: Sequentially assigned IP addresses--why not?
% On the other hand even doing % nothing will be a problem - we appear to have resumed exponential % growth of the routing system again, presumably as multi-homing at % the edges starts to be more and more common. % %Geoff Huston As predicted back in the cidr development days. people multihome for a number of reasons, not the least of which is the avoidance of a single point of failure. Turning the routing system into the functional equivalent of monoply pyramid will bring down the rath of the regulators, even if there is "nothing" we can do. -- --bill
Sequentially assigned IP addresses--why not?
Seriously, As was pointed out recently, IPV6 will croak much sooner than it needs to for the simple reason that we structure routing intelligence into the address assignment. Wouldn't it be better by far, to assign new addresses from 000...1, and map to routing information however we may code it? The memory and processor steps required would be trivial compared to the agony of running out of space again. I'm sure this was argued before. But, it seems to me that the wrong direction has been taken. Gordie Corzine Compaq Global Services (but not speaking for Compaq)
Re: Sequentially assigned IP addresses--why not?
"Corzine, Gordie" wrote: Seriously, As was pointed out recently, IPV6 will croak much sooner than it needs to for the simple reason that we structure routing intelligence into the address assignment. This is some sort of urban legend. If a routeable prefix was given to every human, using a predicted world population of 11 billion, we would consume about 0.004% of the total IPv6 address space. (The actual calculation is 11*10^9/2^48 since there are 48 bits in an IPv6 routing prefix. Or 11,000,000,000 / 281,474,976,710,656 = 0.39 ) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Brian E Carpenter Program Director, Internet Standards Technology, IBM On assignment for IBM at http://www.iCAIR.org Board Chairman, Internet Society http://www.isoc.org Non-IBM email: [EMAIL PROTECTED]
RE: Sequentially assigned IP addresses--why not?
Using the IP address, you index into a table with 100 M entries, pick up an index into the 75K entry routing table. You now have two tables that require maintenance, that's all. If customer changes ISP, their entry in the first table is changed. Link is down, the second table's mechanisms handle it. Use a 64 bit processor architecture, memory is cheap. Re-architecting the Internet is going to become all but impossible. Its a matter of separating routing from identification. Look, my days as an engineer are a distant memory, so I won't try to work this out in detail. Maybe there are irrefutable reasons why this can't be done, but I do believe the current architecture will lead to premature exhaustion of the address space. Gordie From: Steven M. Bellovin Wouldn't it be better by far, to assign new addresses from 000...1, and map to routing information however we may code it? The memory and processor steps required would be trivial compared to the agony of running out of space again. The problem is that we (as a profession) don't know how to do that. We have to make routing scale, and that demands aggregation, which in turn demands structured addresses. Look at it this way. We have about 75K routes in the "default-free zone" now. If we just assigned addresses sequentially, we'd need a route for every endpoint. There are what, 100,000,000 nodes today, and more tomorrow? We can't handle 3 orders of magnitude increase in the size of that table, let alone what it will be in a few years.
Re: Sequentially assigned IP addresses--why not?
In message [EMAIL PROTECTED], "Cor zine, Gordie" writes: Using the IP address, you index into a table with 100 M entries, pick up an index into the 75K entry routing table. You now have two tables that require maintenance, that's all. If customer changes ISP, their entry in the first table is changed. Link is down, the second table's mechanisms handle it. Use a 64 bit processor architecture, memory is cheap. Re-architecting the Internet is going to become all but impossible. The issue isn't table lookup; it's the routing table calculation (and, in the case of your particular example, the sheer amount of data that has to be passed around). Put another way, how does each router know what should be in those 100M entries? Its a matter of separating routing from identification. Phrased somewhat differently, there are a lot of people who agree, though it's still a controversial notion. See if you can find a copy of draft-ietf-ipngwg-esd-analysis-06.txt (or -05) -- it's a description of the best worked-out proposal, plus a refutation of it. (I disagree with the refutation, but I'm not going to go into that now -- I think that the proposal is sound.) Briefly, the idea is to use the high-order 8 bytes of the v6 address for inter-site routing, and the low-order 8 bytes for host id.) But that still requires hierarchical assignment and routing for the high-order 8 bytes. *No one* knows how to do it any differently. Look, my days as an engineer are a distant memory, so I won't try to work this out in detail. Mere assertions that it is possible, in the face of the prevailing wisdom that it isn't, just won't cut it. Maybe you're right, maybe it can be done -- and if so, it won't be the first time that the accepted wisdom is wrong. But the Maybe there are irrefutable reasons why this can't be done, but I do believe the current architecture will lead to premature exhaustion of the address space. Apart from the fact that 128 bits is Really Big, v6 is supposed to have easy renumbering, so that we can renumber sites as they're move around to different pieces of the topology. --Steve Bellovin
Re: Sequentially assigned IP addresses--why not?
Brian Carpenter writes: This is some sort of urban legend. If a routeable prefix was given to every human, using a predicted world population of 11 billion, we would consume about 0.004% of the total IPv6 address space. Surely you recall the quotation attributed to Thomas J. Watson: "The world will never need more than five computers."
Re: Sequentially assigned IP addresses--why not?
The problem is that we (as a profession) don't know how to do that. We have to make routing scale, and that demands aggregation, which in turn demands structured addresses. The telephone company figured out how to avoid problems decades ago. Why the computer industry has to rediscover things the hard way mystifies me.
Re: Sequentially assigned IP addresses--why not?
*No one* knows how to do it any differently. I have an idea: Let's merge IP addresses with telephone numbers. A person will have one IP address for each telephone number he owns, and vice versa, and the two numbers will be the same. Because the identifying number of a telephone is open-ended at both the front and back of the number, there is no limit to the number of addresses that can be accommodated, and the addresses can be used for routing without any danger of exhausting the address space. Example: My machine would be, say, .85794... to people on my block ..44785794... to people in my city ...37744785794... to people in my state .1737744785794... to people outside my country ...421737744785794... to people on Mars .401. to other machines on my home LAN .4015 to the subnet of machine 402 on my home LAN The digits in common between the two machines are not explicitly specified. The address space extends to infinity in both directions. The addressing scheme would locate the starting digit and the number of significant digits, so my full address would be any of the following: 32768-4-4015 32768-3-401 32744-00015-421737744785794 32767-8-94015487 The scheme would allow for starting digits and lengths in excess of 1-65534. The starting digit would be a plus or minus offset, allowing infinite expansion in either direction (there would be no root, but there would be a level 0). I want to talk to a machine in Zumbalu. It's address is 32744-00016-4216849200420283: ...4216849200420283.. Zumbalu ...4217377447857940.. me ...xxx6849200420283.. strip out common digits I connect to 32747-00013-6849200420283. I want to talk to my next-door neighbor. Her address is 32740-00020-04754217377447858662: 04754217377447858662. Jane 04754217377447857940. me 8662. strip out common digits I connect to 32766-4-8662. I have three physical routes from my machine; I select the one with the highest starting digit that is equal to or lower than the start digit of my destination address: R1 = 32768 = nope, too high R2 = 32740 = OK R3 = 32000 = too low Obviously other details can be worked out. This is just back-of-envelope stuff. The important thing is that there is unlimited room for expansion. Additionally, individual nodes in the network need only really know about their immediate neighbors. You wouldn't need worldwide root servers or anything like that.