Contents I. Introduction
Currently, the Internet technology lacks appropriate techniques and features to provide dependable services needed for many critical applications. These applications include, but are not limited to, health care, safety, transportation and multimedia, where compliance with quality of service (QoS) requirements (e.g., bandwidth, delay, jitter and loss) must be guaranteed over time. This motivated the research community towards innovative mechanisms aiming, among others, at advanced Internet resource allocation to guarantee QoS requirements with differentiation to ease service convergence over the Internet. Traditionally, QoS control approaches deploy resource allocation by changing the system configurations (WFQ) on a per-flow basis, that is, whenever a new flow is to be established. Hence, a session setup is subject to QoS reservation, usually by resorting to the Resource Reservation Protocol (RSVP) [1] to coordinate the operations in nodes along communication paths. Although per-flow solutions guarantee QoS, they lack scalability due to excessive states and signaling exchanges, which grow with the number of flow requests and jeopardize performance with undesired consumption of CPU, energy and memory. In view of this, Class of Service (CoS) based QoS architectures (e.g., Differentiated Service (DiffServ) [2] and Multiprotocol Label Switching (MPLS) [3]) are known as scalable alternative for control aggregation. However, while class-based control driven by per-flow reservation signaling [4] reduces state overhead, its undue signaling frequency can easily overwhelm devices I/O interfaces.
