SDN-Assisted Performance-Oriented Bandwidth Allocation Mechanisms
Join Zoom Meeting
https://nus-sg.zoom.us/j/91955282382?pwd=QTlpb0tmVXdwVzlsbWJ1dHNxNDdHZz09
Meeting ID: 919 5528 2382
Password: 650678
Abstract:
More recently, the rapid development of the Software-Defined Networking (SDN) has fostered network management by enabling a logically centralized network control-plane to have direct control of the network data-plane. Enabled by SDN, in this thesis we investigate the design space of bandwidth allocation, that is traditionally controlled by TCP (Transmission Control Protocol) at the end-host and look into how the network can make a careful bandwidth allocation to better support the performance requirements of upper-layer entities (e.g., applications and service policy). We then demonstrate how SDN-based bandwidth allocation can further: 1) capture and improve the performance of stream analytics application, 2) provide service differentiation with performance guarantees based on service operator requirements.
In the first part of this thesis, we present a measurement study that reveals impairments with today's default TCP in the data-intensive stream analytics. We design bandwidth allocation algorithms for assisting stream analytics in achieving better performance during the runtime. Then, we implement an App-aware, a prototype cross-layer bandwidth allocation framework based on a popular open-source distributed stream processing platform, Apache Storm, together with the OpenDaylight controller. Extensive experiments are carried out with real-world analytical workloads on top of a realistic SDN-enabled fat-tree network topology. The results validate the effectiveness and efficiency of our proposed framework and algorithms.
In the second part of the thesis, we propose DiffPerf, a class-based service architecture for the Internet access networks. At a macroscopic level, DiffPerf dynamically allocates bandwidth to service classes (comprising user traffic flows), which enables access providers (APs) to trade-off fairness, and at a microscopic level statistically differentiates and isolates user flows to help them achieve fairer performance, alleviating a severe RTT (Round Trip Time) unfairness problem. We implement DiffPerf on the OpenDaylight controller and evaluate it for Dynamic Adaptive Streaming over HTTP (DASH) application on a large-scale testbed of a cluster of machines interconnected using an SDN switch. Our evaluations demonstrate the practicality and flexibility of DiffPerf in providing differentiated services to users and maintaining a fair quality level for those of the same service class.
Our findings suggest that proposed mechanisms for network-assisted bandwidth allocation actively improve the performance of the upper applications, make the network fairer and robust, and accommodate more diverse users.