Russ Hobby is the Director of Advanced Networked and Scietific Applications at UC Davis. He has been a part of the Internet 2 Project and has served on the Archetecture Committee and the Engineering Committee. He is also involved with the Corsortium for Educational Network Inititives in California (CENIC) where he serves as the chair of the Technical Planning Group. CENIC is the organization that is coordinating Internet 2 for institutions in California. Russ Hobby, Director UC Davis Information Technology Advanced Networked and Scientific Applications Davis, CA 95616 email: rdhobby@ucdavis.edu phone: (916)752-0236 fax: (916)752-0220 --------------------------------------------------------------------------- An Accounting Model for a Multiple Levels of Internet Service Russ Hobby, UC Davis The current Internet has one level of service, best effort. However there has been considerable talk about the need for multiple levels of service and Quality of Service (QoS) on the Internet. It has been recognized that multiple levels of service requires a means for resource control to prevent everyone from simply selecting the best service. Resource control usually translates to charging money, but in some closed environments units other than dollars may be used to allocate resources. Regardless of the units, some means of accounting is necessary to determine the resources allocated and consumed by the end user. This paper briefly reviews traditional communications accounting models and presents a new one. Methods of accounting may not seem that technical, but in truth they will set much of the framework for the Next Generation Internet technology. The Internet has traditionally used flat charges based on the size of the communications ěpipeî. This has been attractive to customers since the costs are easy to predict. It also has not had financial disincentives for use which has allowed the rapid growth of the Internet. Recent developments in the commercial Internet have refined this model somewhat by charging for peak usage over a period of time. This refinement has allowed the installation of a large physical pipe, but it charges for the maximum virtual pipe used. This model still allows predictable charges and also allows quick change to demand by opening up the virtual pipe. However, this modified model and the original Internet model break down with the introduction of multiple levels of service and the need for resource control to the end user. Another model to account for communications services is the one used by telephone systems. Here the end user is charged for each instance of use. While the phone system has one level of QoS, it does offer various types of services, each of which must be accounted in its use. Thus this model could be used for Internet accounting with multiple levels of service. However, a good portion of the computing power to operate the phone system is consumed by accounting and billing. All details of each call must be compiled by all participants in servicing the call. It is desirable in any model to keep the resources required for accounting to a minimum. The two models above are the extremes. The Internet model maintains minimal accounting information but can not track multiple levels of service. The Telephone model maintains information on everything, but at great accounting cost. We will now look at a model between the two. In this model information on service used by an organization is collected by the organization at the desired level of detail. For example, if an organization wants to track service information for each end user, it would could do so. However, external service providers for the organization would maintain accounting information at the organization level. For example, the organization would receive a bill from an external service provider for bulk service at each service level. The organization could then use its own detailed information to allocate costs to individual end users. The organizationís detailed information could be maintained for all traffic on the network, internal and external, or the organization may choose to just collect information on traffic to external providers at network exit points. This local control of accounting detail allows an organization to implement resource control or billing policy in a way that best suits the organization. This model simplifies accounting for the service providers since they do not have to track each end user but can aggregate all accounting information for an organization. However, much of the traffic on the Internet transits multiple service providers and each service provider will want their fair share when providing an end-to-end service. One way is to use the method above for organizations at a higher level of hierarchy. Organizations are now like end users and service provider are like the organization. At an exchange point between service providers, a service provider will account for services used by each of its customer organizations, but will receive a settlement bill from the peer service provider for bulk services that pass through the exchange point. In this way a service provider does not have to maintain accounting information for peer providerís customers. Since traffic for any service on the Internet is two way, there remains the question of who is the requester of the service and how will the reverse traffic be accounted. In the current Internet, traffic in each direction between two devices is treated independently. Routing for each direction may be via different paths. Since this is unlikely to change in the near future, accounting for each direction must also be treated independently. It has been argued that this would be unfair to information providers to pay for the reverse traffic when a requester of the information wants high priority network services. The network accounting model described here would treat this as an application problem and require the application to have a mechanism to charge back the requester for costs of the high priority network services. This is not the most desirable solution. One would prefer let the requester deal with all network charges directly. However this would require much more accounting on the network to identify traffic in both directions for a particular connection. In summary, the accounting method above puts the burden of collecting the details on those that need to have the details for resource control or billing. It allows those not interested in the details to aggregate accounting information and thus reduce the computational burden. Obviously further work is required to fully define the model. This work will have to be done in conjunction with the development of the Internet QoS model.