Brief Announcement: Bridging the Theory-Practice Gap in Multi-commodity Flow Routing

Introduction

In the concurrent multi-commodity flow problem, we are given a capacitated network G = (V,E) of switches V connected by links E, and a set of commodities ${\cal K} = \{(s_i,t_i,d_i)\}$. The objective is to maximize the minimum fraction λ of any demand d _i that is routed from source s _i to target t _i . This problem has been studied extensively by the theoretical computer science community in the sequential model (e.g., [4]) and in distributed models (e.g., [2,3]). Solutions in the networking systems community also fall into these models (e.g., [1,6,5]), yet none of them use the state-of-the-art algorithms above. Why the gap between theory and practice? This work seeks to answer and resolve this question. We argue that existing theoretical models are ill-suited for real networks (§2) and propose a new distributed model that better captures their requirements (§3). We have developed optimal algorithms in this model for data center networks (§4); making these algorithms practical requires a novel use of programmable hardware switches. A solution for general networks poses an intriguing open problem.