Principle of Autonomous Decentralized Flow Control and Layered Structure of Network Control with Respect to Time Scales

Network control mechanisms in networks can be categorized by the time scales on which an individual control takes effect and they form a layered structure with respect to the time scales. In high-speed networks, the processing speed of each node is high but the propagation delay is the same as in slower networks, because it is constrained by the speed of light. So, the framework of network control on a short time scale should be an autonomous decentralized system [1]. We have proposed an autonomous decentralized flow control mechanism applicable to high-speed networks [2], and have investigated its stability and adaptability to dynamic network state [3]. In this article, to clarify the aim of our flow control mechanism, we show supplemental explanations from the following three points of view: • Layered structure of network control with respect to time scale and the inevitability of autonomous decentralized control on a short time scale. • The relationship between local decision making at a node and the performance of the whole network, in order to achieve high global performance as a micromacro link. • The principle of our flow control derived from the efficient framework of parallel computing as an actual approach to realize appropriate global performance by controlling local decision making.