SRA@XX.LCS.MIT.EDU (Rob Austein) (04/21/88)
Hang on, this is getting a little too far down in the bits. Let's ignore the implementation details (counting packets, counting PSNs, etc) for the moment, and concentrate on what this whole business is designed to accomplish: 1) Generate some revenue to fund (and expand?) the operational network, and 2) Provide some negative feedback (that's a technical term, not a normative statement) so that users will make "efficient" use of the available network resources. Let us further assume, that (1) will take care of itself (with the help of the bean counters who are going to be fixing the rates anyway) and that our job is to decide what kind of things really need to be charged back so that (2) will be both fair and effective. I submit that, at the moment, the two most critical resources on the Internet are: a) Long distance trunk bandwidth, in particular (but not limited to) the three transcontinental ARPANET trunks, and b) Gateway CPU time (level-3 routers), in particular (but not limited to) the core gateways and ARPANET <=> MILNET "mailbridges". I would add ARPANET/MILNET PSN overhead, but that's beating a dead horse, no offense to the martyrs at BBN who keep that code running. These are also, of course, the two worst-imaginable places to try to intsert accounting telemetry, precisely because they are so overloaded. Nevertheless, if we are going to have to start paying money per usage for some resource, I for one would prefer to be paying for these. Perhaps some of the income can be used to increase the numbers of trunks and core gateways until they can adaquately handle the load. The point someone made a few days ago about examining the incentive implications of a policy before implementing it is well taken. For example, I'd hate to see the above musings turned into a general charge for router CPU time that encouraged everybody to connect networks with level-2 bridges. I applaud the idea of getting a Real Economist to analyze this mess so that we can giggle at his-or-her ideas rather than each other's. Lastly, if we're going to use highway analogies, let's not forget US 1A southbound through the Sumner Tunnel to Boston, where the traffic routinely backs up so badly, in spite of the toll, that it's -faster- to go 10 miles out of the way on state highways and city streets than to take the direct route. Much of the traffic is people who are either paying by the mile (taxis from Logan Airport) or people who are using the Tunnel simply because they don't know any other route. It happens that just last week I found myself TELNETing from MIT to a site in CT (to which we have a high speed three-hop link via NSFNet) via the ARPANET to Pittsburg and whoknowshow from there because the routing software couldn't tell which route was faster. I'm not sure how to implement a penalty charge for stupid software, but it'd probably be a good thing if we could; picture vendor-of-your-choice being told that ever single customer site would be hit with a surcharge until they fixed their broken software! People who think this is a joke should remember Paul Mockapetris's claim that a bug in bind v4.5 was eating up about a KL-10's worth of CPU time on the root domain servers last July. --Rob
nelson@sun.soe.clarkson.edu (Russ Nelson) (04/21/88)
In article <SRA.12391999638.BABYL@XX.LCS.MIT.EDU> SRA@XX.LCS.MIT.EDU (Rob Austein) writes, and I edit: ... >2) Provide some negative feedback (that's a technical term, not a > normative statement) so that users will make "efficient" use of > the available network resources. ... >I submit that, at the moment, the two most critical resources on the ... > Perhaps some of the income can be used to increase the >numbers of trunks and core gateways until they can adaquately handle >the load. Another possibility: Charge users heavily for the use of the bottlenecks (whatever they may be at the time). Use the resultant income to FIX the bottlenecks. Theoretically (and in a perfect world :-) this will result in a network in which the load is spread evenly and there are no bottlenecks. Implications: o Short term bottlenecks due to down equipment should not be charged for, unless they recur, at which point they become long term bottlenecks. o The cost model is apparent to anyone who wants to do some pinging. Of course, they'll pay for their curiosity. o A site could easily run up a big bill unless the accounting is done in a timely manner. Without timely negative feedback, you can get into oscillations. o This still doesn't address who pays for which packets, just the amount charged for each packet. o Maybe a nominal fee for each packet to cover general costs? -- char *reply-to-russ(int network) { if(network == BITNET) return "NELSON@CLUTX"; else return "nelson@clutx.clarkson.edu"; }
sean@cadre.dsl.PITTSBURGH.EDU (Sean McLinden) (04/22/88)
The biggest single argument against chargeback is that is does NOT take into account the value of the network as a substrate for some greater activity. In terms of the volume of business on the DARPA side of the Internet, I suspect that the biggest users, who are, most probably, the biggest contributors to the evolution of the network, are hardly the deepest pockets, and this includes the many universities and research groups that provide software such as the recent TCP/IP updates to Unix, without charge to the users. There is no fair way to measure this contribution (unless we put it to a quorum), or to predict from where will the next significant contribution come. If the network were an end in and of itself, a chargeback policy based on usage would be quiet reasonable. Under a Libertarian system of government it would also make sense, but many other programs aimed at a greater good are not billed on the basis of usage but on the basis that the good to society represents a greater "fairness" than billing on a usage basis. A previous author, in criticizing my highway analogy, actually argued in favor of my point by noting that toll road was less utilized than the alternate routes which were subsidized by taxation rather than usage tax (except indirectly, by gasoline taxes). The arguement here is that there is a societal good which justifies the expenditure of public funds without regard to direct usage. In the sense that this benefits all of us, independent of our ability to pay, one might argue that it is more benevolent than a system which is more "fair". The problem with many policy-makers is that they love to deal with numbers and go to great pains to quantitize any problem so that they can deal with it using well established (if unimaginative), numeric systems. There is a qualitative issue here, which deals with our economic and technologic competitive edge as a function of information sharing, which cannot easily be reduced to a scalar. We have an obligation to study that thoroughly before we institute the policies of some pencil pusher who needs to show some federal budgeteer a black line that puts them in the clear and forces the rest of us to deal within the limits of short sighted policy-makers. I agree that a study should be done, but participants should not be limited solely to knife wiedling accountants and the clever technocrats who have demonstrated that we have the technology to institute such as system as packet accounting, but also to those people who are able to see how our future as developers and users of this technology and as a nation, would be best served. Sean McLinden Decision Systems Laboratory University of Pittsburgh
perry@MCL.UNISYS.COM (Dennis Perry) (04/22/88)
Sean, I agree with you arguments whole heartedly! Your argument is essentially that the communications cabpability provided by networks like the Arpanet provide a higher level good than just commuications. This is often thought of as 'infrastructure' in some circles, where the infrastructure is what allows society as a whole to operate at a higher level than it could otherwise. How does one pay for infrastructure? Usually thru some means other than usage, although that could be a part of the rate structure. More often infrastructure is paid for thru taxes of some form. Taxes, as we all know, are also often used to provide societal gains, and not all pay according to use (perhaps inversely? :-). I suspect that usage charges on the Arpanet will cause unknown sociological behavior which may turn out to be worse for the government than for the research community as a whole, and what started out as a way to solve a simple budgeting problem may well end up creating a more complicated infrastructure problems which will work against DARPA's desire to foster scientific research in the US and aid the competitiveness of other countries. dennis
perry@MCL.UNISYS.COM (Dennis Perry) (04/23/88)
Barry, In New Mexico we had a saying (and I am sure it applys to other states as well) that one should vote early and often! :-) I agree with suggestions made by some that a study of the possible outcomes of varioius policies on charging would be an interesting thing to do. Such a study could help convince those that make such decisions that certain policies might not be worth pursuing. After all, many decisions get made because of 'hot buttons' that might be better avoided than pushed. I would be glad to organize a group of people that would like to meet in DC or whereever to discuss such a study, make up an agenda, parcel out assignments, and the reassemble to put the paper together. dennis
bzs@BU-CS.BU.EDU (Barry Shein) (04/23/88)
The way I've heard it is: Vote Often, Vote Early, Vote for James Michael Curly! (Curly was a notorious ward-healer who became mayor of Boston in the 1930's I believe, they still talk about those "good ole days" when corruption went to the -right- people.) Well, at least it scans... I would be interested in attending a meeting in DC (or wherever) to discuss such issues. Assembling an invitees list would be interesting although for starters it might just need "activists", others can be contacted once several of us figure out how to start saying the same things. I believe this upcoming election might make this an opportune time to start, a new administration might be more receptive to new-sounding programs such as computer networking. It would be hard to imagine any broad-based objection to it (other than those of a particularly privatizing bent, which should be addressed.) -B
philipp@LARRY.MCRCIM.MCGILL.EDU (Philip Prindeville [CC]) (04/24/88)
The way I've heard it is: Vote Often, Vote Early, Vote for James Michael Curly! (Curly was a notorious ward-healer who became mayor of Boston in the 1930's I believe, they still talk about those "good ole days" when corruption went to the -right- people.) Well, at least it scans... Is Curly the one who was re-elected while doing time because of a federal conviction? Something to do with putting too many friends and relatives on the city payroll? (Maybe that was Hurley) Ah, Boston... I believe this upcoming election might make this an opportune time to start, a new administration might be more receptive to new-sounding programs such as computer networking. It would be hard to imagine any broad-based objection to it (other than those of a particularly privatizing bent, which should be addressed.) This is a good point. And from the sound of it, you might already have certain such groups in mind? -Philip
CERF@A.ISI.EDU (04/25/88)
Dennis, we pay a tax on gasoline which goes to maintain the Interstae Highway system. Gas usage is related to road usage - though how much of the use is on interstantes and how much on local roads isn't clear cut. I still think a usage-related charge is preferable to the fully subsidized free good we have today; there isn't any negative feedback on use, so we have congestion. Vint
bzs@BU-CS.BU.EDU (Barry Shein) (04/25/88)
This is coming down to "do we form bread lines, or do we bake more bread?" Some of it is a futures issue, given a reasonable "taxation" scheme could we keep ahead (or reasonably in line with) the traffic or do we consign ourselves to needing negative feedback to control usage? Even the Interstate Highway system recognizes that you need more highways going in/out of NYC than Coldwater Flats, which would seem to help. This is a two-edged sword, you justify more highways to NYC because there's more gas taxes collected there, but you still have the freedom to have a highway through Coldwater Flats (a tributary or exit or whatever makes sense) even if they could never justify such a think given a fair-share payback scheme. Maybe that's part of the problem, how do you deal with a few researchers who are deserving (that is, would get access at a larger institution) but won't generate enough packet charges to justify (payback) a net connection, the volume clout is just not there? Leave them out in the cold? Tell them to request special and specific subsidy from their funding source? Can some of that be reflected in the bandwidth they and others get? I think part of the idea of infrastructure includes subsidizing otherwise unprofitable ventures, like a highway exit to Coldwater Flats, on the assumption that most everyone needs minimal access, while still being able to respond appropriately to large needs. -Barry Shein, Boston University
CERF@A.ISI.EDU (04/25/88)
Barry (Shein): The reason I like the notion of trying to find a charge-back scheme is that it puts some motivation for efficient use into the loop. For cases where the need for or utility of service exceeds revenues generated, it is possible to subsidize (like lifeline service on the telephone system). I like that because it makes the subsidy visible and forces a decision about providing the subsidy to those who need it. It would be nice to discover that at some point these services could be provided commercially at affordable rates so that the system need not be run by the government at all. At the moment, I have the feeling that commercial rates would be prohibitive - but if there is an economy of scale, it might be that the commercial rates could be reduced if the entire Internet traffic were added to existing commercial traffic. I haven't looked at that at all recently and we'd need some statistics (which is why working on charge-back schemes is good - we may learn enough to figure out make the system pay for itself). Vint
perry@MCL.UNISYS.COM (Dennis Perry) (04/25/88)
Vint, I belive that you gasoline tax illustrate the point I was trying to make, namely, usage charges are ignored by those who can afford it, e.g. they buy low efficiency automobiles like laborginis. It is not clear how much charging it takes to make a difference. Even with HOV in norther VA on I66 and other roads, many, if not most, people still prefer to drive one to an automobile. I posit that usage charges, to affect behvaior, will have to be unreasonably high. And then when they affect behavior, that behavior will become 'antisocial', that is, they will not use the system they way that is optimal, so efficiency becomes an issue again. My experience says that port charges, connect time, etc are not unreaonable type of charges. Volume usage charges are counter productive and drive people to find other means of solving their problems. After all, how much communications does it take to do the research you are involved in. Do you stop your research just because you cannot afford to pay the network charges? I suspect what will happen is you will pick up the telephone and play telephone tag with someone, a much less efficient way to communicate your research ideas. Or you may resort to no communication and just publish your results without feedback from you peers. Both of these alternatives are 'antisocial' behvior in that it is the opposite behavior expected from developing the net in the first place. But you might say, what about users who are using resources attached to the net? Again, if you talk to the operators of those resources, you will find that most of them do not care about the net except to provide service to their customers. They can provide better direct service then a general purpose net can, and probably cheaper. And the efficiency issue arrises here as well. If volume sensive charging were in effect, users of supercomputer centers may well ask for printouts to be done at the center and mailed to the user in order to avoid large charges for printouts. This results in time delay and increased cost to the research, where the cost in this case is the time it takes to turn around a compuation from one set of inputs to another. Again, connect time is one thing, volume sensitive charging is another. When ISDN becomes fully functional, it may become feasible to build a high speed network that is build around circuit switches instead of packet switches and one will get rid of the idea of dedicated lines. You just dial up what you need, use it for the time you need it, and then hang up. Just like the telephone service today, you do not get charged for how fast you talk (volume of data), but how long you talk (time) plus some type of port charge (basic monthly fee). In dedicated line systems, like the Arpanet, etc., there is no contention for a port in current implementations. Perhaps the gateways or PSNs could refuse virtual circuit connections based upon load so that connect time has some value associated with it, such as some level of service. The issue is not an easy one, but I don't think one should run down the road without exploring the issues. The DoD is already experiencing people moving off the DDN because of the expense required for the service provided. Many are setting up their own networks, because the commong network does not work well for them. What are they using? Well, the Navy is setting up a UUCP type of network based on dial up lines and 9.6 kbit/s service! It isolates them from other, it is inefficient, etc., but it was done because of perceived problems with upcoming usage charges on Milnet and performance issues that such charging would generate. enough for now, dennis
perry@MCL.UNISYS.COM (Dennis Perry) (04/25/88)
Vint, one of the reasons that DARPA is haveing problems with paying for the Arpanet, is the charging mechanism which is in place. There are many connections which would like to pay for their port charge, but cannot because DARPA has no way currently to collect the money. So we dismantle the infrastructure because of an accounting problem. Usage charges are based on the flawed idea that one can change behavior based upon how much it costs. All it really does is inforce behavior based upon how much money you have. Can you really say that someones research is more valuable because they have lots of grant money to pay for communication charges? dennis
haverty@CCV.BBN.COM (04/25/88)
It's encouraging to see a discussion like this happening. I think it may be useful and/or important to note the distinction between cost-recovery and "cost"-feedback. All of a network's costs must be recovered somehow, and mechanisms must be in place to collect the data about costs to support collecting the needed funds. Basing that scheme on end-user usage is only one scheme, which is in favor right now - the analogy on the interstates might be if every vehicle's odometer reading was reported to some accounting system that produced bills. Cost feedback is a separable question, though closely related. The goal is to encourage efficiency and reduce waste usually, but also may be to encourage certain kinds of use that the benevolent network owner wants to promote. Feedback may be in the form of inconvenience (slow network performance, long gas lines at the pumps), or money, or availability (a T1 circuit to your campus, an interstate interchange at your town). My suspicion is that a mix of such techniques is needed. As people have noted, a blanket usage-based scheme might tend to stifle new ideas or uses; a free-networking approach drives costs out of sight for the benefactor. Maybe we need a scheme which promotes new ideas, but assures that efficiency gets introduced along the way to large scale usage of any particular idea? This sounds like a topic which needs wide input and thinking. Does it make sense to hold a session of some kind at the upcoming TCP/IP conference in Santa Clara this summer? Dan Lynch's shows seem to be drawing a good mix of academic, industry, government, and user representation. Dennis Perry - maybe in addition to pulling together that paper you mentioned, you could pull together a session to present results and a large dose of open discussion? Jack
sam@ftp.COM (Shelli Meyers) (04/25/88)
In article <8804222230.AA09291@bu-cs.bu.edu>, bzs@BU-CS.BU.EDU (Barry Shein) writes: > > I believe this upcoming election might make this an opportune time to > start, a new administration might be more receptive to new-sounding > programs such as computer networking. There is an interesting article in the April 18th issue of _Computerworld_ regarding the high tech issues playing a part in this year's election, and the specific stands of Bush, Dukakis, Gore, and Jackson on things like federal tax credit for R&D, telecommunications regulations, corporate tax rates, etc. The article is somewhat brief and pretty general, but worth a look. Shelli Meyers FTP Software, Inc.
ahill@CC7.BBN.COM ("Alan R. Hill") (04/26/88)
Why is it that I have a fear that worthy discussions of issues such as these will not survive the introduction of the tarrif? Alan
hine@comp.vuw.ac.nz (John Hine) (04/27/88)
This has been an interesting discussion. Unfortunately it seems to ignore those of us a long way from the centre of activity. COnsider the case for usage charges. I would have to agree that usage charges don't always create an immediate increase in efficiency of use, but they might introduce some fairness. Down here we pay real money for our X.25 link to UUNET, and we pay both ways. Because the Internet has no chargeback mechanism we pay for you to send us mail. This makes an e-mail connection expensive but still worthwhile. The problem arises when someone decides they have something you would like and, without thinking, pops it in the e-mail. Voila, we have a $60 bill for something that might not have been solicited. We've had worse cases than this and it happens with annoying regularity. Some form of chageback would at least get these people to think about their actions -- at least after their first $60 bill. For a lot of things air mail is still a cost effective communication medium! jh -- ----------------------------------------------------------------------- Domain: hine@comp.vuw.ac.nz Dept. of Computer Science UUCP: ...!uunet!vuwcomp!hine Victoria University P.O. Box 600, Wellington, NZ
perry@MCL.UNISYS.COM (Dennis Perry) (04/27/88)
Alan, certainly if I had to pay for each character I sent I would try to find a better way of participating in this discussion, or not participate at all. After all, this conversation is not related to very many of our jobs, but is extreamly important in that the network facilitates the performance of our job. dennis
CERF@A.ISI.EDU (04/27/88)
Dennis, I don't see much difference between connect time charge and volume if the data rate is fixed (e.g. at 64 kb/s). Some people would argue that connect time is regressive if you are charged the same to send a little as a person who sends a lot. That's one reason the Telenets and Compuserves charge more for 1200 baud than for 300 baud access. Vint
perry@MCL.UNISYS.COM (Dennis Perry) (04/27/88)
Vint, we had different connect time charges at Los Alamos, depending on the rate, so rate sensitivity is was understood by the users. The difference between volume and connect charges is of course rather small, but I think goes the right way to support an infrastructure. After all, we are not arguing about wheather one should pay, just about how one goes about collecting the money and how much for what services. You pay for the resourse you use in keeping others out of the system. The resourse you use to get to the system, telecommunications, may in fact be dedicated to you, but there were limited number of users who could effectively use the ports available to get to the supercomputers. Now, you could sign on and use the facility at any rate you chose, i.e. usage rate is not the same as bandwidth. Typing characters at 20 word per minute is not the same as transmitting each character at 56 kbit/s. Those who only type text could get by with 2400 to 4800 b/sec service, while those who generated lots of graphics to a textronix would much prefer a 9.6 kbit/s service or higher (we had some that ran substantially higher, I don't remember now how high). In the arpanet one does not normally have the option of connection and disconnecting to the PSN, thus, one has a static connection. What we do have is random receipt and sending of packets accros the interface. Even today, the Arpanet folks have the option of giving you a 9.6 kbit/s line connection to the PSN or a 56 kbit/s line connection. I do not remember for sure, but I believe that the cost to DARPA is the same. This is one of the flawed aspects of current billing for the Arpanet. The only cost to the user is the connection line to the PSN (excpet for early connection which were still being paid for by DARPA, but were being looked at to have the user pay the connect charges.) I am not sure where this conversation is going, but I sense an attitude by some that usage sensivitive charging is the way to go with out looking at alternatives and resultant possible reactions by the community or the purpose of fostering the network in the first place (policy). My bottom line is that the existing system may not be designed to support certain types of charges and one should examine that as well. Do we want to change the protocols, do we have to, how does it affect the policy if we do, etc.? I do not claim to know the answeres, nor do I believe a simple solution is necessarily available outside of the Government to continue to provide 'free' service. Another solution might be for the Government to find a way for connectees to pay their port charge instead of DARPA having to pay it. That way DARPA could continue to subsidize those whom they wished, others would pay. I suspect that solution alone would reduce the DARPA part of the bill to less than a $1M. By the way, I am not convinced that DARPA really wants to reduce its cost that much. The DARPA default connection is to the Milnet, which is going to usage charges. It used to be the Arpanet, but we switched over when the Arpanet became so congested and DARPA folks could not get to ISI to read their mail. I suggested that they switch back, but that has not happened (yet?). dennis
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (04/28/88)
In article <8804262145.AA13529@LANAI.MCL.UNISYS.COM> perry@MCL.UNISYS.COM (Dennis Perry) writes: >Alan, certainly if I had to pay for each character I sent I would >try to find a better way of participating in this discussion, or >not participate at all. If we mst al py per char, ther my b wys to rduc th no o chars xmitd. kwe, BU
mckee@corwin.ccs.northeastern.EDU (George McKee) (04/28/88)
Alternative-Subject: "what price broadcasts?" Dennis Perry <perry@mcl.unisys.com> writes: >... >The resourse you use to get to the system, telecommunications, may in >fact be dedicated to you, but there were limited number of users >who could effectively use the ports available to get to the supercomputers. >Now, you could sign on and use the facility at any rate you chose, i.e. >usage rate is not the same as bandwidth. Typing characters at 20 word >per minute is not the same as transmitting each character at 56 kbit/s. >Those who only type text could get by with 2400 to 4800 b/sec service, >while those who generated lots of graphics to a textronix would much >prefer a 9.6 kbit/s service or higher (we had some that ran substantially >higher, I don't remember now how high). >... >dennis When I think about network service, this statement is more like the mental model I have than the other one that seems to be implicit in much of this discussion. The fact that mail is perhaps the most visible form of network use leads people think about billing in terms of per-message (i.e. per-packet) charges, just like the post office or the telephone company. An alternative model that I think is just as valid is cable television, where you pay (in advance, even) for a guarantee of a certain amount of bandwidth. Subscribers are offered different levels of service, and it is provided whether your TV is on or not. Few people watch HBO 24 hours a day, but that's what you pay for. I'd guess that there's at least as much internet traffic generated by "broadcast" lists such as this one, as there is traffic generated by individual-to-individual messages, though I have no idea how the relevant data could be collected. In a research-oriented community, the free flow of information is widely acknowledged to be essential for productivity. It seems to me that bandwidth limits would be much less painful to run up against than would usage limits. I wouldn't like to be in the middle of transferring some important file and suddenly get hit by a "sorry: packet quota exceeded" message. Or to be told that I've exceeded my network budget with a whole month left in the fiscal year. Bandwidth limits could avoid this kind of trouble because the costs are fixed in advance. Of course, there would have to be ways of limiting bandwith (and other forms of net access, as well) by software rather than hardware, so that senior researchers could obtain the bandwith to transfer their supercomputer output, while undergraduates or other peons who happen to have signons on the same machine/subnet could be restricted to a harmless trickle, but this should be preferable to flooding the network with accounting packets. If this alternative has already been considered and rejected, I apologize. I'd appreciate knowing the reasons, though, if someone could point to an article where this is discussed. - George McKee Software Coordinator College of Computer Science Northeastern University, Boston 02115 CSnet: mckee@Corwin.CCS.Northeastern.EDU Phone: (617) 437-5204 Usenet: the signal/noise ration here is already too low...
LIVINGSTONE@BERT.DECNet (LIVINGSTONE) (04/28/88)
SIGNOFF TCPIP-L UNSUBSCRIBE TCPIP-L