学术文献库

To meet the timing requirements of interactive applications, the no-frills congestion-agnostic transport protocols like UDP are increasingly deployed side-by-side in the same network with congestion-responsive TCP. In cloud platforms, even though the computation and storage is totally virtualized, they lack a true virtualization mechanism for the network (i.e., the underlying data centers networks). The impact of such lack of isolation services, may result into frequent outages (for some applications) when such diverse traffics contend for the small buffers in the commodity switches used in data centers. In this paper, we explore the design space of a simple, practical and transport-agnostic scheme to enable a scalable and flexible end-to-end congestion control in data centers. Then, we present the the shortcomings of coupling the monitoring and control of congestion in the conventional system and discuss how a Software-Defined Network (SDN) would provide an appealing alternative to circumvent the problems of the conventional system. The two systems implements a software-based congestion control mechanisms that perform monitoring, control decisions and traffic control enforcement functions. Both systems are designed with a major assumption that the applications (or transport protocols) are non-cooperative with the system, ultimately making it deployable in existing data centers without any service disruption or hardware upgrade. Both systems are implemented and evaluated via simulation in NS2 as well as real-life small-scale test-bed deployment and experiments.