How many QoS classes are optimal?

This paper presents a novel method to manage the operational risk of noncompliance with respect to network service-level agreements. This method answers the question of how many quality-of-service (QoS) classes a provider shall offer to maximize his return for a given risk and vice versa. We prove that two QoS classes are often optimal for a single QoS parameter and specific networks, namely, single links and some broadcast networks. It is interesting that the non-compliance risk can be managed by turning it into a service feature. ©1999 ACM.