| P&P has its roots in the TDM Voice networks, where it can be defined and implemented in a natural setting. Due to the hop-by-hop nature of IP networks, definition and implementation of P&P requires additional care. One of the main challenges is to keep the overall P&P architecture simple enough so that the management and operations of the networks do not become overly cumbersome, and that the end-to-end P&P requirements are satisfied across multiple autonomous domains. To this goal, the DoD QoS Working Group has defined DoD wide use of the Differentiated Services (DS) code points to indicate the P&P level and the class of service associated with each packet. In general, the use of the DS field is a necessary tool for P&P implementation, but more is needed. If preemption is based only on the DS field, this action runs the risk of dropping packets from multiple users, as opposed to minimizing the number of affected users from preemption. Similarly, when users of the same priority level face congestion, use of the DS field alone cannot ensure that a minimum number of users are affected by the actions taken due to congestion.
Flow based signaling mechanisms such as RSVP can overcome the shortcomings faced when the DS field alone is used for P&P purposes. But RSVP’s scalability problems in backbone networks are well documented in RFC 2208. Thus, although RSVP may be a useful tool within the access networks with limited number of users, it may not be applicable in backbone networks.
An alternative method is to treat the packets according to their flows, in addition to the DS field. Advances in ASIC technology make it feasible to recognize, and keep the state the flows at each router, without making use of signaling methods. Flows are recognized based on attributes such as source and destination IP addresses, and port numbers. In addition, other attributes such as packet sizes and packet inter-arrival times make it feasible to recognize the nature of the flows, e.g., VoIP, and thus allow for the reservation of the corresponding network resources without using signaling protocols. This paper will describe and analyze how flows can be recognized without a priori signaling, and how this information can be used for P&P implementations.
|