Chapter 3: Nonsynchronized Nonhomogeneous AIMD
M. Corless, Christopher King, R. Shorten, F. Wirth · Society for Industrial and Applied Mathematics eBooks · 2016
The previous chapter provided a comprehensive treatment of synchronized homogeneous AIMD. Recall that in synchronized AIMD, all users decrease at every event, and that in homogeneous AIMD, each user employs a single fixed growth rate parameter αi and decrease parameter βi at every capacity event. We now turn our attention to situations in which not all users respond to capacity events in unison and where some users change their growth rate parameter or decrease parameter at different events. An AIMD algorithm in which not all users respond to all events is referred to as nonsynchronized AIMD. This scenario may arise because only some of the users are notified of each event, as in TCP applications, or it may arise because some users do not respond to the notification but instead continue in their increase phase. An AIMD algorithm in which some users change their growth rate parameter αi or their decrease parameter βi will be referred to as nonhomogeneous AIMD. In this chapter, we begin the treatment of synchronized nonhomogeneous AIMD, nonsynchronized homogeneous AIMD, and nonsynchronized nonhomogeneous AIMD. As we shall see, these three cases can be treated in a common framework. We will present results for the deterministic case and introduce some tools that can be used in later chapters in the stochastic case.