17.4.1999
Jyri Lindström
Department of Computer Science and Engineering
Helsinki University of Technology
jplindst@cc.hut.fi
These days the Internet is used to transfer larger amounts of data than previously and requirements for Quality of Service are growing too when interactive multimedia applications are becoming more common. Bandwidth is sometimes a bottleneck in the Internet. That's why there is need for Quality of Service. To use QoS there must be a way to charge customers for usage of common resources according QoS level they are using. This paper is about accounting and billing in the Internet today and in the future.
2.1 Flat fee pricing
2.3 Edge pricing
2.5 the Internet mall
3 Accounting and billing currently in the Internet
3.4 Accounting
4 Accounting and billing in the future
4.1 Sima
4.2 A Framework for Policy-based Admission Control
4.3 Billing
When the Internet was created backbone was funded by Universities that used it. They shared the expenses equally and every body paid for the connection to the Internet not for traffic they generated. Today when the Internet is used widely around the world and is more business oriented companies offering the Internet connections are interested of moving from flat fee pricing to pricing where people would pay for resources they consume. This would rationalize the usage of network resources. With fixed payment customer can use the Internet as much as he wants no matter how much he causes expenses to his ISP (the Internet Service Provider) .
New services like video conferencing and different kinds of multimedia applications consume network resources so much that all the traffic in the Internet will slow down when network get congested. There is two way to resolve from congestion: Buy more bandwidth or implement accounting and billing system that will encourage people to use network resources more sparingly. This paper will present the current situation of accounting and billing in the Internet and some future possibilities.
Flat fee pricing scheme is probably the most used pricing scheme in the Internet today. It means that you pay each month certain fixed amount of money for your the Internet connection and can use it as much as you want without any extra expenses. Monthly payment depends only on the speed of the line.
There are several advantage to the flat fee pricing scheme. First of all it is simple and doesn't have any overhead of the accounting and billing. Because it is fixed customers always know how much their surfing is going to cost in advance. This pricing system helps free flow of information because you do not have to pay any more for information you share. Also posting lists will disappear if they should pay for every single mail that they send.
There are also some disadvantages in flat fee pricing too. There is no priority in flow of the information. When the Internet becomes congested the traffic slows down, IP packets are delayed or even lost. Those who really need the Internet in some productive way are not going to have any faster connection than some average surfer that visits in the playboys-website.
The idea behind usage-based pricing is that costs of the Internet should reflect to amount of usage. User will be billed according to amount of data he has moved, the quality of service he has received and length of the transmission. That way anybody who think he needs a bigger share of the resources of the Internet can have it by paying more.
Good thing in this pricing system is that everybody pays according resources he consumes. It wouldn't be fare if the people who uses the Internet for E-mail and telnet would pay the same amount that people who uses video conferencing tools over the Internet. Network applications consumes more and more bandwidth but when you have to pay for every byte transferred over the Internet it might lead to more economically usage of the Internet too.
Accounting and billing in usage-based pricing definitely will cause overhead in lines and routers and adds costs that the Internet users have to pay. Other disadvantage is that usage-based pricing limits the free flow of information in posting lists. And home pages distribute information that wouldn't be shared if people should pay when somebody downloads it.
Edge pricing system [2] is similar to flat fee pricing and usage-based pricing systems except that all pricing decisions are made at the edge of the Internet (ISP) locally. Edge pricing system allows more flexible and advanced pricing policies such as peak-load could be priced twice as expensive as stable load. Or set some limits that up to 1MB per day is free but up to that it costs 1 mk/MB. Price could also differ according to time of day or congestion of line.
The advantages of edge pricing systems are that they can be taken into use by one ISP at a time. So it doesn't require whole the Internet to support edge pricing scheme at once. Local ISP can decide the policy used to bill customers so that it covers the expenses of providing the Internet connection to them. And pricing system can be made simpler that some global usage-based pricing.
Accounting in edge pricing will cause a lot of overhead especially when using complicated pricing schemes compared to usage-based or flat fee pricing systems. Basically we are billing per packet basis but accounting must be done in more general level because it would cost too much to log every packet send/received to/from the Internet.
Smart market pricing [1] means that user includes with packet he send over the Internet 'bid' how much he is willing to pay for delivery of that packet. When network is congested packets with lowest bids will be delayed or dropped if queues are full. Billing is made according to lowest bid transported. This means that everybody pays as little as possible and still get decent service. If user do not want to pay any extra for lower delay or assured transport he can use the Internet as long as there is no congestion in the path of the IP packet.
Advantages in smart market are that consumer pays always the right price for the service (the Internet connection) if nobody else is using the Internet he can have all the capacity he wants with no additional cost. But if there are too many users using same resources then those willing to pay most can continue using net like before and others who are paying less have to wait.
Disadvantage in smart market is that accounting and billing of traffic can cause much overhead since it requires lots of logic in router to priories IP packets and calculate the correct price for every packet routed.
One possible way to finance the growth of the Internet is Faulhaber's vision [4]. According to Faulhaber the Internet could be like mall today so that companies that use the Internet in their business would offer their customers free access to the Internet. And as a "side effect" they would receive a lots of new customers. In the United states a site alladvantage.com offers you money for the time you surf at the Internet and are looking the banners of the sponsors on your screen while surfing. Maybe in the future the Internet could work in the same way as commercial TV channels. Collecting money by selling advertises.
Most of the ISDN and modem connections to the Internet are priced using flat fee pricing policy. I think that biggest reason is that then ISPs don't need to take care of the accounting. Price is usually very cheap like 50 mk a month. Most of the home users are happy because expenses are known in advance. Addition to fixed price customer have to pay for connection to ISP according normal time-based tariff to his local telephone operator. But as long as phone connections are local price for that is reasonable too. Some ISPs offer the Internet connections that do not require any agreement between the ISP and the customer. Basically it works like normal the Internet connection with ISDN or modem but user pays according to the connection duration. If usage is very minimal then this could be cheaper alternative than fixer monthly payment. But when usage exceeds couple of hour a month it will be more expensive than flat fee pricing.
This is quite new way to connect to the Internet through cable TV line. In Helsinki it is available only small areas in the central of the town. Its speed is good and HTV is saying that they are going to reserve 0,5 - 1 Mbps bandwidth per client downlink. Uplink speed is slower but that is not a problem in normal the Internet usage unless user wants to have a server. Pricing is fixed monthly fee that in HTV is 245mk a month. I think it is quite reasonable price since you have unlimited fast access to the Internet.
Leased lines are usually used by companies that need fast the Internet connections to be able to offer employees opportunity to use the Internet resources in work. Or if company have www-server for homepage and for some other services to their customers they need also leased line unless they are going to use some kind of web-hotel solution offered by almost every ISP. Leased lines are often priced using flat fee pricing. But some ISPs use flat fee prices for traffic to Finland and then additionally accounts the traffic to outside of Finland and prices that according the amount of data transferred to/from other countries.
I phoned both Sonera and Eunet to compare their leased lines services and pricing policies. I asked costs and service quality for 2 Mbps line. Sonera offers this with fixed price of 15990 mk per month that covers all the costs and assured 256 kbps bandwidth to other countries. If customer wants more assured bandwidth to other countries it costs 2600 mk per month per 32 kbps. Eunet's offer is a lot of cheaper 2190 mk including 200 MB traffic to other countries. Additional traffic to other countries costs 495 mk per 100 MB. After 400 MB customer starts to have discounts starting from 10 % to 70 % when traffic to other countries exceeds 15 GB. There is no assured bandwidth to other countries but probably they try to offer enough bandwidth because the have very much money for that traffic. Price doesn't include physical line to Eunet and that costs about 2000 mk per month with standard modem line and might be more if customer want to use some other data transfer media like FASTNET. Eunet basic packet cost total little over 4000 mk. But it will be much more when using resources from other countries.
When traffic out of Finland is between 3000 and 4000 MB the costs of Sonera's and Eunet's the Internet connections are more or less equal. If traffic to other countries is more than 4000 MB then it is clearly cheaper to use Sonera's services. 3 - 4 GB traffic to foreign countries is not even very much because most of the information programs, service packs and other interesting stuff comes from other countries than Finland. When there is many people sharing this connection 4 GB is not much. And when using flat fee pricing there will not be unpleasant surprises when there is more than expected amounts of traffic to foreign countries.
There is lots of different accounting protocol available today. But they are often quite complicated and that's why they are not use in large scale. Eunet's way to account foreign traffic is record traffic amount between IP addresses. They use Cisco routers feature that generates records about IP flows [3]. That data is transferred to accounting machine using telnet connection. Accounting machine uses IP addresses to divide traffic to Finland and other countries. Foreign traffic is then assigned to certain customers and added it to that customers traffic amount to be billed.
In the future there is growing need for system that could handle different Quality of Service levels and accounting and billing according resources that are used. In this chapter we shall study some architectures and protocols that could be used to account and bill customers the Internet usage in the future. There are lots of requirements for billing system of the next generation of the Internet. It should be compatible with current the Internet so it is possible to use it only by some ISPs while other are still using old systems.
SIMA (Simple Integrated Media Access) is described in the Internet drafts [7,8] and [9] which define the architecture and implementation of SIMA. The basic idea is to use differentiated services [10] to provide Quality of Service in the Internet. All the accounting, billing and decisions about Quality of Service is done in edges of the network as described earlier. The core network routers are simple to ensure scalability. SIMA provides only soft guaranties for delivery of IP packets but it divides the available bandwidth according to customers willingness to pay (NBR).
User have to define NBR (Nominal Bit Rate) and select between real-time and non-real-time service classes before making connection. NBR defines the bandwidth the user want to use. The higher the NBR is the more money user have to pay for it. User can although send data with any bit rate he wants no matter what NBR he has defined but the traffic that exceeds the NBR has higher drop preference. It means that when network is congested IP packets with higher drop precedence will be dropped. NBR and the service class can be selected beforehand when making an agreement with the network operator or could be selected separately before making the connection. SIMA architecture doesn't specify that and it is left for ISP to decide how to organize it. Figure 1 describes the architecture of SIMA.
CE = Customer Equipment
A = Access Node
C = Core Network Node
Figure 1: SIMA Network Architecture
Access Node:
Access node constantly measures the MBR (momentary bit rate) of each customer flow. MBR is calculated using exponentially weighted moving averages. Then MBR and NBR is converted to drop-precedence value using following formula:
where Int(x) is the integer part of x.
This means that when the customer sends data to network IP packets are marked with drop precedence. The higher the current transfer speed is relative to Nominal Bit Rate the higher drop precedence is assigned to packet. The user can optimize the packet loss versus cost by changing the transmission rate. The drop precedence and real-time/non-real-time service class is fitted to 3+1=4 bit and access node packs those bits to TOS (Type Of Service) field in IP packet-header.
Core Network Node:
Network Node handles the discarding packets that cannot be forwarded due network congestion and takes care that real-time service class packets are not kept in queues too long. Figure 2 below describes the packet SBU (Scheduling and Buffering Unit) used in SIMA network node.
Figure 2.
The SBU calculates PLa from queue lengths. Or it could have precalculated table of PLa values to ease real-time processing effort. PLa can have values from 0 (queues almost full) to 6 (no significant queues). If drop precedence (PL) of packet is higher than PLa it is discarded. Otherwise it is forwarded to queue according to service class bit. There are two queues in SIMA network node. One is very small like 10 Kbytes for real-time packets. The other is much bigger like 1 Mbytes or more for non-real-time packets. When router is forwarding packets ahead from queues it always takes packets from real-time queue. Only when real-time queue is empty packets from non-real-time queue is forwarded ahead.
This the Internet-draft [5] describes architecture that can be used to provide Quality of Service and accounting to IP networks. It is based on integrated services [RFC 2211, RFC 2212] and RSVP (Resource ReserVation Protocol) [RFC 2205] but also other protocols for resource reservation and service level definition can be used. This architecture offers better guarantees about Quality of service than in SIMA. But this is also more complicated and the overhead of managing resource reservation is higher too. Figure 3 presents the relationship between PEP (Policy Enforcement Point) and other int-serv components within an RSVP router.
PC = Packet Classifier
PS = Packet Scheduler
PEP = Policy Enforcement Point
PDP = Policy Decision Point
Figure 3.
PEP:
PEP is point that takes care of authenticating and puts the policy in action. It consults the PDP and/or LPDP about policy. Usually first from LPDP for making decisions based on policy elements handled locally and after that it send the query to PDP.
PDP:
PDP makes the decisions about service to be offered to customer according policy. PDP may also consult other PDPs in some circumstances. PDP also takes care of the accounting and usage monitoring for billing and network monitoring purposes.
RSVP:
RSVP handles the actual resource reservation in the network according PEPs decisions. To be able to reserve the resources customer wants there must be that capacity available at every line and router in the way to receiver. If there is routers that doesn't support RSVP in the way to receiver than allocated bandwidth cannot be guaranteed.
Figure 4.
Figure 4 describes the architecture of whole network that has several administrative
domains in the way from sender (S1) to receiver (R1). we can also see that every
administrative domain should have at least one PDP. It can be in separate machine that can
be used by several routers. Or it can also be embedded to router as in router D has. If
router doesn't have its own PDP it should at least have LPDP to make it ease a bit the
delay that query to PDP takes.
The PEP - PDP protocol should be reliable, delays should be small. The protocol should be able to carry opaque objects, support two-way transactions, handle multicast groups and support asynchronous notifications. Current protocols like RADIUS, LDAP or SNMP doesn't fill all the requirements specified. We describe here one protocol that could be used as PEP - PDP protocol.
COPS:
COPS (Common Open Policy Service) is described in the Internet draft [6] and it is designed to work with architecture defined above in communication between PEP - PDP. COPS uses TCP as its transport protocol to assure reliable exchange of messages. It relies on IPSEC [11] for security. It can be optionally used to authenticate and secure the traffic between PEP and PDP instead of IP if security is needed.
Actual billing in either of these architecture is not specified very clearly. In SIMA it is assumed that billing is done according Nominal Bit Rate that customer have to decide when making agreement with ISP. It would be possible to decide NBR when opening connection and do separate billing transaction for every connection opened. Or maybe even better would be to use some default NBR that is used normally but if customer want to increase the NBR for some connection he could do it and pay it later with monthly billing or alternatively pay it using micro payments [12] when opening connection.
Micro payments could also be used in architecture described in chapter 4.2. When doing resource reservation customer could pay for resources with micro payment. Micro payments aren't not in used yet but in the future when banks are offering micro payment services it could be the easiest way to pay small amounts of money for services in the Internet.
I think that in the future there will be some kind of usage-based billing. It is going to take a long time before any system will be in widespread use. The biggest reason is probably that at the moment there is enough capacity in core network and the Internet haven't been congested for a while. But traffic is growing steadily while new users are joining to the Internet and new multimedia applications require more and more from network. Sometimes in the future we will see again congestion and that is the day when ISPs must do something to provide service those customers that are willing to pay for it.
There is one problem with billing that depends on the congestion on the network. What if service provides would intentionally keep capacity so low that everyone must use more expensive service if they want to use the Internet at all. Only competition between ISPs can prevent this scenario from happening.
The SIMA seems to be quite competitive model compared to other architecture described in chapter 4.2. It is simple enough and still offers soft Quality of Service that even works. It doesn't require too big changes to core network routers and can work even with standard router. Maybe newer routers could be used after firmware update to work as SIMA network router.
| [1] | MacKie-Mason J. K. & Varian H. R., Pricing the Internet,
10.2.1994 < http://www.spp.umich.edu/courses/744/docs/Pricing_the_the Internet.pdf > |
| [2] | Shenker S. & Clark D. & Estrin D. & Herzog S., Pricing in
Computer Networks: Reshaping the Research Agenda, Release date unknown < ftp://parcftp.xerox.com/pub/net-research/picn.ps > |
| [3] | Cisco Systems, Cisco Billing Architecture, 1999 < http://www.cisco.com/warp/public/779/servpro/solutions/csm/acct/cba_wp.htm > |
| [4] | Faulhaber G. R., Pricing the Net: What Economists Do,
17.3.1995 < http://www.spp.umich.edu/courses/744/docs/stanpapr.pdf > |
| [5] | Yavatkar R., A Framework for Policy-based Admission Control,
the Internet Draft, November 1998 < http://www.ietf.org/the Internet-drafts/draft-ietf-rap-framework-02.txt > |
| [6] | Boyle J. & Cohen R. & Durham D. & Herzog S. & Rajan R.
& Sastry A., The COPS protocol, the Internet Draft, 24.2.1999 < http://www.ietf.org/the Internet-drafts/draft-ietf-rap-cops-06.txt > |
| [7] | Kilkki K., Simple Integrated Media Access, the Internet Draft, June 1997 < http://www-nrc.nokia.com/sima/sima.txt > |
| [8] | Loukola M. & Ruutu J. & Kilkki K., Dynamic RT/NRT PHB Group,
the Internet Draft, November 1998 < http://www-nrc.nokia.com/sima/draft-loukola-dynamic-00.txt > |
| [9] | Loukola M. & Engdahl T. & Forsten J., Traffic Measurements of the
SIMA Access Node Implementation, 18.3.1999 < http://www-nrc.nokia.com/sima/sima_an.pdf > |
| [10] | Kilkki K., An introduction into the philosophy of Differentiated Services,
16.3.1998 < http://www-nrc.nokia.com/sima/introfw.htm > |
| [11] | Atkinson R., Security Architecture for the Internet Protocol, RFC1825, August 1995. |
| [12] | IBM, IBM Micro Payments, 1998 < http://www.hrl.il.ibm.com/mpay/ > |
There was still several documents that I didn't refer directly but which gave me some backround information and helped me to see the whole picture.
