ACCCoRD: Architectural Considerations on the placement of Congestion Control Routines and Data – Extended Abstract

At the IETF, there has recently been a number of proposals for separating the identifier and locator nature of IP addresses. Furthermore, many of these proposals aim to provide support for multi-addressing by making the fact that a host may have multiple parallel IP addresses invisible to the layers above. In this paper we discuss the potential consequences of such a change to congestion control algorithms and the sublayering within the IP and transport layers. In the current TCP/IP stack structure, most of the end-to-end state is cleanly packaged at the transport layer, i.e., in the TCP, UDP, SCTP, DCCH, or equivalent data structures. If the suggested identifier / locator split is adopted with multi-addressing support, it becomes necessary to change the way congestion control parameters are represented in a host stack. In particular, we claim that it will be necessary to make a separation between congestion control parameters that are related to a specific path versus parameters that are related to a given flow or connection.