Theory of Flow Ordering for Computer Networks
Year: 2017
Type: preprint
Abstract: The problem of elephant flow detection has been the subject of intense research for the last fifteen years at least. However, previous research presumes the existence of the abstract concept of flow size, for which the current literature has no formal mathematical definition. In this paper, we develop a theoretical framework that leads to the definition of the abstract concept of flow size in connection with the global quality of service (QoS) requirements of a network. Our framework demonstrates that the problem of elephant flow detection is a particular case of the more general problem of mapping groups of flows onto an arbitrary number of network policies. With this view we introduce the concept of partitioned QoS and identify the convexity properties that lead to a formal definition of the concept of flow size. We demonstrate that metric-based approaches found in the existing literature provide a solution to one particular class of network problems, and that a precise definition of flow size can only be constructed starting from the partitioned QoS problem. Further, we show that if a partitioned QoS problem is nested convex as formally defined in this work, then it can be solved in polynomial time.
Source: arXiv (Cornell University)
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2575483588