The growing number of mobile devices and data-intensive applications pose unique challenges for wireless access networks as well as datacenter networks that enable modern cloud-based services. With the enormous increase in volume and complexity of traffic from applications such as video streaming and cloud computing, the interconnection networks have become a major performance bottleneck. In this thesis, we study algorithms and architectures spanning several layers of the networking protocol stack that enable and accelerate novel applications and that are easily deployable and scalable. The design of these algorithms and architectures is motivated by measurements and observations in real world or experimental testbeds.
In the first part of this thesis, we address the challenge of wireless content delivery in crowded areas. We present the AMuSe system, whose objective is to enable scalable and adaptive WiFi multicast. AMuSe is based on accurate receiver feedback and incurs a small control overhead. This feedback information can be used by the multicast sender to optimize multicast service quality, e.g., by dynamically adjusting transmission bitrate. Specifically, we develop an algorithm for dynamic selection of a subset of the multicast receivers as feedback nodes which periodically send information about the channel quality to the multicast sender. Further, we describe the Multicast Dynamic Rate Adaptation (MuDRA) algorithm that utilizes AMuSe's feedback to optimally tune the physical layer multicast rate. MuDRA balances fast adaptation to channel conditions and stability, which is essential for multimedia applications.
We implemented the AMuSe system on the ORBIT testbed and evaluated its performance in large groups with approximately 200 WiFi nodes. Our extensive experiments demonstrate that AMuSe can provide accurate feedback in a dense multicast environment. It outperforms several alternatives even in the case of external interference and changing network conditions. Further, our experimental evaluation of MuDRA on the ORBIT testbed shows that MuDRA outperforms other schemes and supports high throughput multicast flows to hundreds of nodes while meeting quality requirements. As an example application, MuDRA can support multiple high quality video streams, where 90% of the nodes report excellent or very good video quality.
Next, we specifically focus on ensuring high Quality of Experience (QoE) for video streaming over WiFi multicast. We formulate the problem of joint adaptation of multicast transmission rate and video rate for ensuring high video QoE as a utility maximization problem and propose an online control algorithm called DYVR which is based on Lyapunov optimization techniques. We evaluated the performance of DYVR through analysis, simulations, and experiments using a testbed composed of Android devices and o the shelf APs. Our evaluation shows that DYVR can ensure high video rates while guaranteeing a low but acceptable number of segment losses, buffer underflows, and video rate switches.
We leverage the lessons learnt from AMuSe for WiFi to address the performance issues with LTE evolved Multimedia Broadcast/Multicast Service (eMBMS). We present the Dynamic Monitoring (DyMo) system which provides low-overhead and real-time feedback about eMBMS performance. DyMo employs eMBMS for broadcasting instructions which indicate the reporting rates as a function of the observed Quality of Service (QoS) for each UE. This simple feedback mechanism collects very limited QoS reports which can be used for network optimization. We evaluated the performance of DyMo analytically and via simulations. DyMo infers the optimal eMBMS settings with extremely low overhead, while meeting strict QoS requirements under different UE mobility patterns and presence of network component failures.
In the second part of the thesis, we study datacenter networks which are key enablers of the end-user applications such as video streaming and storage. Datacenter applications such as distributed file systems, one-to-many virtual machine migrations, and large-scale data processing involve bulk multicast flows. We propose a hardware and software system for enabling physical layer optical multicast in datacenter networks using passive optical splitters. We built a prototype and developed a simulation environment to evaluate the performance of the system for bulk multicasting. Our evaluation shows that the optical multicast architecture can achieve higher throughput and lower latency than IP multicast and peer-to-peer multicast schemes with lower switching energy consumption.
Finally, we study the problem of congestion control in datacenter networks. Quantized Congestion Control (QCN), a switch-supported standard, utilizes direct multi-bit feedback from the network for hardware rate limiting. Although QCN has been shown to be fast-reacting and effective, being a Layer-2 technology limits its adoption in IP-routed Layer 3 datacenters. We address several design challenges to overcome QCN feedback's Layer- 2 limitation and use it to design window-based congestion control (QCN-CC) and load balancing (QCN-LB) schemes. Our extensive simulations, based on real world workloads, demonstrate the advantages of explicit, multi-bit congestion feedback, especially in a typical environment where intra-datacenter traffic with short Round Trip Times (RTT: tens of s) run in conjunction with web-facing traffic with long RTTs (tens of milliseconds).
Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8GB2GH3 |
Date | January 2017 |
Creators | Gupta, Varun |
Source Sets | Columbia University |
Language | English |
Detected Language | English |
Type | Theses |
Page generated in 0.0018 seconds