New network technologies and communications protocols are being developed to address the needs of multiple quality-of-service applications. There is now a critical need to design efficient and feasible mechanisms for allocating the various service qualities in an integrated services network.
I describe a mechanism for efficient allocation in an integrated services network. The mechanism elicits truthful revelation of the strength of user preferences for network resources, and implements a Pareto efficient allocation: that is, no alternative allocation could make any user better off without making at least one user worse off.
The mechanism is appropriate for a network that schedules resources in advance. This architecture permits an allocation mechanism that is responsive to the state of demand. State-responsive allocation allows the "rationing rule" to assign the available resources to the highest value users during congested periods. My network model captures multiple qualities of service, while retaining the simplicity of a single service dimension (guaranteed bandwidth) through reliance on recent results on ``effective bandwidth'' as a way to collapse different service requirements into a single measure of required bandwidth.