Ref: Your note of Fri, 15 Apr 1994 10:52:00 -0400 Henry,
This assumes alot, including customers that don't switch providers and inject more specific routes (this is happening already).
No, I assumed that most (but NOT all) of the customers who switch providers would renumber. I can further weaken this assumption by allowing aggregation above the immediate provider level (you can do this with CIDR).
You're also assuming that we have a mechanism for aggregating routes for folks who choose not to (or can't, for some reason).
That is not so far-fetched assumption. Proxy aggregation is likely to happen, whether some folks like it or not.
CIDR only buys us a *little* time..
Remember there is "CIDR movie", but there is also "CIDR book". You need to consider both... Yakov.
a bit less than 28,000 currently configured "Internet" network numbers believe they have permanently gained their class B's and C's and will NEVER give them up and renumber. Or at least the ones in the US believe that. a bit less than all assigned network numbers total believe that they have gained their class B's and C's and will never give up and renumber. You have provided no incentive (carrot) for individual companies to do the right thing. CIDR is a perfectly engineered suspension bridge unconnected to either bank of the river. Marty
Ref: Your note of Fri, 15 Apr 1994 10:52:00 -0400
Henry,
This assumes alot, including customers that don't switch providers and inject more specific routes (this is happening already).
No, I assumed that most (but NOT all) of the customers who switch providers would renumber. I can further weaken this assumption by allowing aggregation above the immediate provider level (you can do this with CIDR).
You're also assuming that we have a mechanism for aggregating routes for folks who choose not to (or can't, for some reason).
That is not so far-fetched assumption. Proxy aggregation is likely to happen, whether some folks like it or not.
CIDR only buys us a *little* time..
Remember there is "CIDR movie", but there is also "CIDR book". You need to consider both...
Yakov.
Marty,
a bit less than 28,000 currently configured "Internet" network numbres believe they have permanently gained their class B's and C's. Or at least the ones in the US believe that.
a bit less than all assigned network numbers total believe that they have gained their class B's and C's and will never give up and renumber
You have provided no incentive (carrot) for individual companies to do the right thing.
Let us try to answer your question with another question: Do you want a routable large scale global Internet ? It is hard to imagine supporting a truly huge Internet without relying on hierarchical routing (CIDR is simply a realization of hierarchical routing). And if you do plan to rely on hierarchical routing, then you need to understand how to deal with the issue of containing address entropy (due to switching among providers) without renumbering. It seems naive and perhaps irresponsible to think about flat routing (based on network numbers). It should be a goal to make this renumbering simple. We'd like to suggest that folks with alternative proposals to CIDR should put their alternative proposals on a table and explain, among other things, how their proposals would be deployed and used and how these proposals would be better than CIDR. Hitting the right time frame turns out to count! When people got network numbers in the past they were getting addresses for the research Internet. It is important to understand that the research Internet was a great thing, but we are now working on the global public Internet and we desperately needed new routing and addressing systems. We should establish that we are in a transition from the research Internet to the global public Internet and we subsequently can not just use uncoordinated IP addresses and still have a workable system. This is not dissimilar to what happened when local phone exchanges started to get interconnected during the advent of long distance telephone services. There needs to be a globally coordinated address space to make this work. Reasoning by analogy with the phone system is a powerful argument. People change phone numbers all the time, they don't absolutely revolt because the phone system is so valuable. Some elect to get 700 numbers, but they *PAY* for this service. We suggest the following subjects be carefully considered: The old addresses of the research Internet need to be reorganized into the global public Internet addressing plan which is based on CIDR. Those addresses not currently globally routed will not be routed. These new customers of the Internet should get their addresses from their immediate providers. (This could be softened if there is a commitment by the customer to enter into the transition ASAP). This also would cover the case of provider switching under CIDR. Those addresses that are currently routed will *eventually* be migrated to CIDR allocations. This may take some time, on the order of years (2-5). We could look for the simple cases first (small/tiny sites). It is not fair to get people to renumber when they attach to the Internet when they see that people already attached are just sitting pretty. We need to be consistent in the application of standards and rules. Marty has brought up the subject of a carrot: The carrot is getting global Internet routing. The stick is not getting global Internet routing. It is a dull and boring argument, but it is the core of the debate. There is extreme value in what we are trying to build with the global public Internet, and we need to impress on the customer base that we need their help to make it possible to achieve our goals. We are not saying this is going to be easy, but it is rare that something worth having comes for free. Peter & Yakov P.S. The number of uncoordinated IP addresses is higher than 30K.
No. I do not want a routable large scale global Internet. I want a mixed mode Information SuperHighway which is a mixture of routing and layer-two switching. In fact I have that right now extensively internally, and increasingly externally. I believe that the the routed form will have some hierarchy, but it will NOT be exclusively hierarchial. PSI did hierarchial routing in 1992 internally and transparently to everyone. In fact at the current scale of the routed form there is no need for hierarchy architecturally. You have an IMPLEMENTATION problem in the NSFNet backbone and some other "key" places which will not accept any larger routing tables. What if you used cisco's with 64Megs of memory or 128Megs? Would we be playing "henny penny" right now if that was true? Should we really be considering a routing architecture imposition at 100,000 routes or 200,000 routes? The implications of CIDR are extensive and have not been "sold" to the world, there is no carrot and thinking that you can impose all hierarchial routing on everyone is niave. I think your metaphor of phone numbers is wrong, I think the metaphor is portable 1.800 imposed by the FCC and embraced by everyone who used it. You are going in the opposite direction and are imposing non-portable 1800 implicitly with CIDR, it will not play in Peoria. Sign up for temporary compression of the routing table and then growth beyond the current volume some number of months later. Forget renumbering, you are a party theorist and your cadre does not represent the entire soviet. You REPRESENT the research Internet taking a position on the way the global public Internet is going to growing up is absurd. You've no experience on what is happening in the real world of providing the service to end customers. There are lots of issues you have not given any creedance to even if you have given them limited consideration. But nothing in your messages leads me to believe that you have given them consideration. I'll bet we could have flat routing if less than 100 places did some upgrades, and then we could keep some of the benefits of flat routing. PSI is doing CIDR, but we are NOT signed up for some of the draconian theories you and others are espousing. so Forget - renumbering (ready to deal with Solomon's or GE's lawyers?) - not accepting individual network numbers (or Columbia's?) - continued growth (you will compress, and you will slow, you will not stop) We will have to incent (carrot) people to do the right thing, you have no plan (predictable government style), we need to come up with one, and it will then take 12 to 18 months to have material affect. Marty
Let us try to answer your question with another question:
Do you want a routable large scale global Internet ?
It is hard to imagine supporting a truly huge Internet without relying on hierarchical routing (CIDR is simply a realization of hierarchical routing).
And if you do plan to rely on hierarchical routing, then you need to understand how to deal with the issue of containing address entropy (due to switching among providers) without renumbering. It seems naive and perhaps irresponsible to think about flat routing (based on network numbers). It should be a goal to make this renumbering simple.
We'd like to suggest that folks with alternative proposals to CIDR should put their alternative proposals on a table and explain, among other things, how their proposals would be deployed and used and how these proposals would be better than CIDR. Hitting the right time frame turns out to count!
When people got network numbers in the past they were getting addresses for the research Internet. It is important to understand that the research Internet was a great thing, but we are now working on the global public Internet and we desperately needed new routing and addressing systems. We should establish that we are in a transition from the research Internet to the global public Internet and we subsequently can not just use uncoordinated IP addresses and still have a workable system. This is not dissimilar to what happened when local phone exchanges started to get interconnected during the advent of long distance telephone services. There needs to be a globally coordinated address space to make this work. Reasoning by analogy with the phone system is a powerful argument. People change phone numbers all the time, they don't absolutely revolt because the phone system is so valuable. Some elect to get 700 numbers, but they *PAY* for this service.
We suggest the following subjects be carefully considered:
The old addresses of the research Internet need to be reorganized into the global public Internet addressing plan which is based on CIDR.
Those addresses not currently globally routed will not be routed. These new customers of the Internet should get their addresses from their immediate providers. (This could be softened if there is a commitment by the customer to enter into the transition ASAP). This also would cover the case of provider switching under CIDR.
Those addresses that are currently routed will *eventually* be migrated to CIDR allocations. This may take some time, on the order of years (2-5). We could look for the simple cases first (small/tiny sites).
It is not fair to get people to renumber when they attach to the Internet when they see that people already attached are just sitting pretty. We need to be consistent in the application of standards and rules.
Marty has brought up the subject of a carrot:
The carrot is getting global Internet routing.
The stick is not getting global Internet routing.
It is a dull and boring argument, but it is the core of the debate. There is extreme value in what we are trying to build with the global public Internet, and we need to impress on the customer base that we need their help to make it possible to achieve our goals.
We are not saying this is going to be easy, but it is rare that something worth having comes for free.
Peter & Yakov
P.S. The number of uncoordinated IP addresses is higher than 30K.
I tell ya, fellas, Marty is right. Renumbering is a looser. People will give us more trouble than its worth it. E.g., PSI would never go for giving up their Class A, as it means too much to them. Must help them milking the cows. Lets just bandaid the current addressing structure and protocols as best we can for a little longer, and under the premise that we have to jettison the IP address space anyway and completely and each and every bit a little while down the road. We should have gone for CLNP years ago (didn't Clark suggest that in, uh, '86 or so already)? If only for the sole benefit of a forced renumbering by means of a brand new and hierarchical addressing space called NSAPs. Who cares about petty details like the protocols chosen, and some bells'n'whistles nonsense in each packet. The address is what counts. Color and size of the envelope doesn't matter. Right on, Marty!
PSI is easy (when compared to the below) I can't imagine going back to 5000 seperate companies and say "Give me your network number, by such and such a date, or I turn you off" I'll be glad to refer their lawyers to Peter, LLNL, the FNC or anyone who has a big wallet/purse/money-bag-under-bed to settle the issue. That is a non-deterministic path, it will not be ruled by fiat. A clean slate as below (tounge in cheek or not) is the chance to create the perfect information state (make sure all transactions are checkpointed through the FinCen etc). And of course there are other possabilities for clean slates. (Possabilities not probabilities btw). Marty
I tell ya, fellas, Marty is right. Renumbering is a looser. People will give us more trouble than its worth it. E.g., PSI would never go for giving up their Class A, as it means too much to them. Must help them milking the cows. Lets just bandaid the current addressing structure and protocols as best we can for a little longer, and under the premise that we have to jettison the IP address space anyway and completely and each and every bit a little while down the road. We should have gone for CLNP years ago (didn't Clark suggest that in, uh, '86 or so already)? If only for the sole benefit of a forced renumbering by means of a brand new and hierarchical addressing space called NSAPs. Who cares about petty details like the protocols chosen, and some bells'n'whistles nonsense in each packet. The address is what counts. Color and size of the envelope doesn't matter. Right on, Marty!
marty, If the return on routing entry (RORE) is large enough no one is going to particularly upset if mega-corp sends in a routing entry. The place to get gain is at the tiny places (my home for example with a subnet with a tiny number of hosts).
"Give me your network number, by such and such a date, or I turn you off"
As I suggested in my note this is too strict a position to take. I suspect that PSI like most providers work with their customers, and could facilitate this kind of transition over time. The issue is to get from the current state to a better state. I am not unwilling to consider larger routers, hierarchy prudently applied, renumbering, etc. Why limit yourself when you have real problems to solve. The implication that I am gratuitously suggesting renumbering without solving a real problem is incorrect. If everything was going just right, the topic would not come up. If we figure out how to do this without any hierarchy as you seem to imply would be desirable, then I hope the vendors build routers big enough to hold routes to all the telephone outlets in China. We need some abstraction, and CIDR simply carries the IP subnetting model out to its logical conclusion. The thought that we can eliminate all hierarchy is as silly as demanding a strictly hierarchical system. Let's not confuse CIDR with renumbering. CIDR is to get reasonable routing handles on "places" in the Internet. In some cases this will be a 30 bit long prefix identifying a single host on a LAN and in some cases it will be the prefix to all the IP systems in Mozambique. Appropriate prefixes to meet routing requirements. I think this fits your notion of a non-deterministic path. cheers, peter
If the return on routing entry (RORE) is large enough no one is going to particularly upset if mega-corp sends in a routing entry. The place to get gain is at the tiny places (my home for example with a subnet with a tiny number of hosts).
We have a solution for that, we call it an InternetRelay, they are part and parcel of what we are deploying on cable. Surprise!
"Give me your network number, by such and such a date, or I turn you off"
As I suggested in my note this is too strict a position to take. I suspect that PSI like most providers work with their customers, and could facilitate this kind of transition over time. The issue is to get from the current state to a better state. I am not unwilling to consider larger routers, hierarchy prudently applied, renumbering, etc. Why limit yourself when you have real problems to solve. The implication that I am gratuitously suggesting renumbering without solving a real problem is incorrect. If everything was going just right, the topic would not come up.
And while your fording the stream with your tanks to liberate those shackled by RevisionistDemocrats you accidently crush ten civilians and your 120mm gun goes off and you take out the town Hospital. "I'm sorry, I'm from the government, I'm here to help". Basically you can cause more problems than you are solving with your current tac.
If we figure out how to do this without any hierarchy as you seem to imply would be desirable, then I hope the vendors build routers big enough to hold routes to all the telephone outlets in China. We need some abstraction and CIDR simply carries the IP subnetting model out to its logical conclusion. The thought that we can eliminate all hierarchy is as silly as demanding a strictly hierarchical system.
I'm not eliminating all hierarchy, I'm arguing (at least) mixed mode, and of course no renumbering.
Let's not confuse CIDR with renumbering. CIDR is to get reasonable routing handles on "places" in the Internet. In some cases this will be a 30 bit long prefix identifying a single host on a LAN and in some cases it will be the prefix to all the IP systems in Mozambique. Appropriate prefixes to meet routing requirements. I think this fits your notion of a non-deterministic path.
If you think the current IP is going to deal with China then you have lost sight of what are reasonable design goals. Start again with new white paper. Don't saddle the current IP with totalitarian-administered CIDR to reach that goal. Marty
You have an IMPLEMENTATION problem in the NSFNet backbone and some other "key" places which will not accept any larger routing tables.
Just to be clear, you are in fact mistaken. The RS/6000 routers are not having problems with the current routing load. They no doubt will have problems some day, you can't keep hardware forever if the task it needs to do is continually growing, but the current problems are with 16M ciscos in environments with relatively rich connectivity. You keep complaining about what is being done wrong, but have said very little about what should be done instead. I'd be much more interested in hearing the latter. Dennis Ferguson
CIDR is a technology similiar to a Meat Cleaver It can be used for good or for ill. We've watched CIDR for a long time, and perceived it as having "stealth" implications in how it would be used. We are now seeing how it will be used, and how it is intended to be used. Some of them sound pretty bad (we are not surprised). We want to see CIDR used for good. Not for secret or not-so-secret agendas which we do not agree with. We want both hierachial and non-hierachial routing to proceed in parallel. Marty
You have an IMPLEMENTATION problem in the NSFNet backbone and some other "key" places which will not accept any larger routing tables.
Just to be clear, you are in fact mistaken. The RS/6000 routers are not having problems with the current routing load. They no doubt will have problems some day, you can't keep hardware forever if the task it needs to do is continually growing, but the current problems are with 16M ciscos in environments with relatively rich connectivity.
You keep complaining about what is being done wrong, but have said very little about what should be done instead. I'd be much more interested in hearing the latter.
Dennis Ferguson
We want to see CIDR used for good. Not for secret or not-so-secret agendas which we do not agree with. We want both hierachial and non-hierachial routing to proceed in parallel. I believe that we have unanimous consent on these items. So, I'll play Suzanne Vega and ask "what's the matter here"? Tony
Marty, In the Internet we believe the issue will be name based portability, not address portability. This level of decoupling will permit the functionality of portable 800 numbers. +1 800 I-LIKE-IP is simply a name which the telephone company maps to something which which they in turn route to. It is important not to confuse naming with routing and packet forwarding. It is also important to note that CIDR is based on mask and match so it is not exclusively hierarchical as you imply. It allows for a mixture of flat and hierarchical routing, as the need arises. This must be true since today the Internet exhibits both flat uncoordinated addresses and hierarchically assigned addresses. We seem to be able to route them in together. CIDR allows you to dial in the level of hierarchy you need. This seems to be sound architecturally. cheers, Comrades Peter and Yakov P.S. I always like the way you color our debates in ideological terms, I really enjoy picturing you in a Jeffersonian Wig. -- pf
In YOUR Internet you believe in name based portability. That is NOT what the people who have those numbers believe. You are confusing the technology vs the reality of the marketplace. That is not to say that you can't turn on your propaganda machine and make everyone believe that Comrade Peter is a friend of all Internet children. CIDR with masks defeats the substantial purpose's that you are selling CIDR to solve, I can't wait to see the processing impact of the increasingly sparse matrix's you are pushing. Not that I have heard anyone buy into it yet though. Marty PS: He was a red head. I think i'm casting myself as Medevev in this context. ---------
Marty,
In the Internet we believe the issue will be name based portability, not address portability. This level of decoupling will permit the functionality of portable 800 numbers. +1 800 I-LIKE-IP is simply a name which the telephone company maps to something which which they in turn route to.
It is important not to confuse naming with routing and packet forwarding.
It is also important to note that CIDR is based on mask and match so it is not exclusively hierarchical as you imply. It allows for a mixture of flat and hierarchical routing, as the need arises. This must be true since today the Internet exhibits both flat uncoordinated addresses and hierarchically assigned addresses. We seem to be able to route them in together.
CIDR allows you to dial in the level of hierarchy you need. This seems to be sound architecturally.
cheers, Comrades Peter and Yakov
P.S. I always like the way you color our debates in ideological terms, I really enjoy picturing you in a Jeffersonian Wig. -- pf
In YOUR Internet you believe in name based portability.
What fraction of Internet users know their DNS names better than they know their IP addresses? What number of people know my email address vrs the number who know my IP address? I bet more users know more URLs than IP addresses.
That is NOT what the people who have those numbers believe.
You are confusing the technology vs the reality of the marketplace.
No, I am considering what the users really want, which is Internet services, and THEY are the ones who don't care about address space, technology, etc.
That is not to say that you can't turn on your propaganda machine and make everyone believe that Comrade Peter is a friend of all Internet children.
I will plead guilty to selling ice cream on a hot day. Sort of a market driven thing. You can scream all you want about how it is going to cool off tonight, I've got to go back to taking care of that line of customers.
CIDR with masks defeats the substantial purpose's that you are selling CIDR to solve, I can't wait to see the processing impact of the increasingly sparse matrix's you are pushing. Not that I have heard anyone buy into it yet though.
CIDR always has masks, by defn. CIDR without masks is like a black russian without Kahlua. If you are refering to the notion that through hierarchical abstraction you lose information, you are certainly correct. However, most networks don't want to know *all* of the gory details of networks halfway across the globe, or even in the next state.
Marty
petrov
PS: He was a red head. I think i'm casting myself as Medevev in this context.
Harder to image, but I'll work on it. Long live the revolution!
What fraction of Internet users know their DNS names better than they know their IP addresses? What number of people know my email address vrs the number who know my IP address? I bet more users know more URLs than IP addresses.
I didn't make my point clear. You are talking about individual users. I'm talking about the decision makers and operators of the networks at individual institutions. They are not going to be easily sold on renumbering or non-portable network numbers. They are going to have to be incented. Marty
Peter & Yakov, the only carrot/stick combination that is going to work is to charge for announcements - period. Make a charging scheme that punishes many small announcements out-of provider block announcements Convince the big transit providers to implement it and it will percolate downwards. This has some problems too but is the only carrot/stick combination that will work better than just convincing people to be good citizens. Daniel Speaking for myself only
"Peter S. Ford" <peter@goshawk.lanl.gov> writes:
Marty,
a bit less than 28,000 currently configured "Internet" network numbres believe they have permanently gained their class B's and C's. Or at least the ones in the US believe that.
a bit less than all assigned network numbers total believe that they have gained their class B's and C's and will never give up and renumber
You have provided no incentive (carrot) for individual companies to do the right thing.
Let us try to answer your question with another question:
Do you want a routable large scale global Internet ?
It is hard to imagine supporting a truly huge Internet without relying on hierarchical routing (CIDR is simply a realization of hierarchical routing).
And if you do plan to rely on hierarchical routing, then you need to understand how to deal with the issue of containing address entropy (due to switching among providers) without renumbering. It seems naive and perhaps irresponsible to think about flat routing (based on network numbers). It should be a goal to make this renumbering simple.
We'd like to suggest that folks with alternative proposals to CIDR should put their alternative proposals on a table and explain, among other things, how their proposals would be deployed and used and how these proposals would be better than CIDR. Hitting the right time frame turns out to count!
When people got network numbers in the past they were getting addresses for the research Internet. It is important to understand that the research Internet was a great thing, but we are now working on the global public Internet and we desperately needed new routing and addressing
systems. We should establish that we are in a transition from the research Internet to the global public Internet and we subsequently can not just use uncoordinated IP addresses and still have a workable system. This is not dissimilar to what happened when local phone exchanges started to get interconnected during the advent of long distance telephone services. There needs to be a globally coordinated address space to make this work. Reasoning by analogy with the phone syste m is a powerful argument. People change phone numbers all the time, they don't absolutely revolt because the phone system is so valuable. Some elect to get 700 numbers, but they *PAY* for this service.
We suggest the following subjects be carefully considered:
The old addresses of the research Internet need to be reorganized into the global public Internet addressing plan which is based on CIDR.
Those addresses not currently globally routed will not be routed. These new customers of the Internet should get their addresses from their immediate providers. (This could be softened if there is a commitment by the customer to enter into the transition ASAP). This also would cover the case of provider switching under CIDR.
Those addresses that are currently routed will *eventually* be migrated to CIDR allocations. This may take some time, on the order of years (2-5). We could look for the simple cases first (small/tiny sites).
It is not fair to get people to renumber when they attach to the Internet when they see that people already attached are just sitting pretty. We need to be consistent in the application of standards and rules.
Marty has brought up the subject of a carrot:
The carrot is getting global Internet routing.
The stick is not getting global Internet routing.
It is a dull and boring argument, but it is the core of the debate. There is extreme value in what we are trying to build with the global publi c Internet, and we need to impress on the customer base that we need their help to make it possible to achieve our goals.
We are not saying this is going to be easy, but it is rare that something worth having comes for free.
Peter & Yakov
P.S. The number of uncoordinated IP addresses is higher than 30K.
Peter,
When people got network numbers in the past they were getting addresses for the research Internet. It is important to understand that the research Internet was a great thing, but we are now working on the global public Internet and we desperately needed new routing and addressing systems. We should establish that we are in a transition from the research Internet to the global public Internet and we subsequently
I don't think this will fly. I was around when we had a "research internet". I even had two class A addresses assigned to me [yes, I know I should have kept them :-) ]. I think that was quite a while ago and does not apply to the majority of internet sites today. Certainly not for all of the sites labeled as commercial under the AUP.
address space to make this work. Reasoning by analogy with the phone system is a powerful argument. People change phone numbers all the time, they don't absolutely revolt because the phone system is so valuable.
The phone number analogy does not apply here. When you phone number changes (and in most cases today it is only the area code, not the local number) the telephone still works. This is not the case for computers and routers. They break. Some hosts may not even boot. When the phone number changes and someone dials the old number the user gets a recording with the new number to dial. We have not built the current internet to do anything like this. Bob
participants (8)
-
Bob.Hinden@Eng.Sun.COM
-
Daniel Karrenberg
-
Dennis Ferguson
-
hwb@upeksa.sdsc.edu
-
Martin Lee Schoffstall
-
Peter S. Ford
-
Tony Li
-
yakov@watson.ibm.com