Xiaojun Hei

Hierarchical content routing in large-scale multimedia content delivery network

Content delivery network (CDN) is an intermediate layer of infrastructure that helps to efficiently deliver the ever increasing multimedia content from content providers to a large community of geographically distributed clients. Content routing is an essential component of CDN architecture. In this paper we propose a hierarchical content routing architecture for large-scale CDN, in which CDN servers perform inter-cluster content routing based on two-level hierarchical overlay network. We analyze the routing overhead and the corresponding CDN performance of different intra-cluster content routing schemes. In particular, we propose a semi-hashing based scheme for intra-cluster content routing and a content-query based scheme for inter-cluster content routing. Through qualitative analysis and simulations we show that the semi-hashing based scheme is scalable (small routing overhead), efficient (high content sharing efficiency), and flexible (adjustable parameters).

Towards low-redundancy push-pull P2P live streaming

P2P live streaming systems are developed in two major approaches: tree-push versus mesh-pull. The hybrid push-pull streaming, as an emerging and promising approach, offers a good tradeoff between traffic overhead and system throughput. In this paper, we demonstrate that video redundancy is a large contributor of the traffic overhead in push-pull systems. To reduce the traffic overhead, we propose simple but effective sub-stream scheduling and re-scheduling mechanisms, implemented in a push-pull streaming prototype called Low-redundancy Streaming (LStreaming). To demonstrate its effectiveness on reducing the traffic overhead, we conduct both simulation and prototype experiments and compare the proposed LStreaming with random mesh-pull and GridMedia. The simulation results show that LStreaming significantly reduces the total traffic overhead, i.e., up to 33% and 37% reduction compared with mesh-pull and GridMedia in dynamic P2P environments, respectively. LStreaming also achieves the throughput, more close to the optimal value than the other two schemes, and sustains a better video playback quality. The prototype experiments show that LStreaming is practical and achieves the expected performance.

Model-based end-to-end available bandwidth inference using queueing analysis

End-to-end available bandwidth estimation between Internet hosts is important to understand network congestion and enhance the performance of Quality-of-Service (QoS) demanding applications. In this paper, we investigate model-based available bandwidth measurement via the use of an active probing stream. A general end-to-end measurement framework, which unifies the current research approaches and highlights insights for measurement practice, is proposed. Within this framework, the end-to-end available bandwidth is inferred based on the measurement of the performance metrics of an active probing stream. We study two probing streams: Poisson and periodic probing. Of particular interest to our investigations is the Squared Coefficient of Variation (SCV) of the inter-probing packet arrival time at the receiver. The performance comparison of the available bandwidth measurements based on loss models and delay models indicates that the delay-based measurement exhibits many advantages over the loss-based measurements, such as accuracy, overhead and robustness. We conducted a comparison study between the proposed SCV-baged probing scheme, namely, SCVProbe, and Pathload using ns-2 simulation in terms of probing accuracy, convergence time and overhead. Our evaluation results indicate that SCVProbe achieves similar or even better measurement accuracy than Pathload with much less probing time and smaller overhead.

Wavelength converter placement in least-load-routing based optical networks using genetic algorithms

Feature Issue onn Next-Generation WDM Network Design and Routing (WDMN). We studyn the problems of routing and wavelength converter placement in optical networks withn sparse wavelength conversion. We propose a new dynamic routing algorithm with twon new path cost functions based on the concept of least-load routing (LLR) with sparsen converter placement, and we discuss the application of genetic algorithms (GAs) ton determine the optimal location of wavelength converters so that the call-blockingn probability is minimized. Simulation results show that the proposed dynamic routingn algorithms perform significantly better than shortest-path (SP) routing andn fixed-alternative routing (FAR) in terms of the call-blocking probability. The GAn model is able to obtain a nearly optimal solution of the wavelength convertern placement problem within a reasonable time, and its performance is better than thatn of two other popular heuristic placement algorithms.

Stochastic Relay Routing in Peer-to-Peer Networks

Network Address Translation (NAT) commonly prevents nodes without globally valid IP addresses from establishing direct Internet paths. In peer-to-peer networks, peers may utilize intermediate nodes as relays for this NAT traversal. We develop a stochastic relay routing algorithm for selecting appropriate relay nodes. The proposed relay routing algorithm is constructed in a stochastic programming framework by leveraging the actual delay of local links and the statistical delay distributions of non-local overlay links. Single or multiple paths are established via relays between two peers for achieving packet delivery with low delay and small loss. The simulation results showed that the proposed stochastic single/multi-path routing algorithm achieved a much lower packet delay than deterministic shortest path algorithms, which utilize average link delays. We demonstrated the effectiveness of the path diversity provided by our algorithm in reducing packet loss significantly via simulations. Our algorithm is fully distributed and requires only accurate local information. The employment of our algorithm is beneficial for provisioning real-time streaming applications in peer-to-peer networks.

Available bandwidth measurement using Poisson probing on the Internet

In this paper, we investigated a non-intrusive probing methodology for available bandwidth measurement based on the analysis of the departure process of an active Poisson probing stream. Unlike the self-congestion based available bandwidth measurement, non intrusive techniques are meant to infer the available bandwidth along a path without congesting the path. We propose to probe the end-to-end path using small size packets with exponentially distributed time between consecutive probing packets. Of particular interest to our investigations, is the squared coefficient of variation (SCV) of the inter-departure process of the probing stream. The Internet is modelled as single server queue with two concurrent streams, the probing traffic stream and the cross traffic, we rely on the results on M/sub 1/ + M/sub 2//GI/sub i//1 queueing system and a heavy traffic approximation model to analyze the departure process of the probing stream. Thus, in a real measurement system, given the measured SCV of the probing stream, inverting the approximation helps inferring the load of the cross traffic on an end-to-end-path.

A light-weight available bandwidth inference methodology in a queueing analysis approach

End-to-end available bandwidth estimation is important in understanding network congestion and enhancing service quality. In this paper, we investigate a light-weight probing method for available bandwidth measurement in a queueing analysis approach. Unlike the self-congestion based measurement approach, a light-weight probing technique infers the available bandwidth along a path without congesting the routers along the path. Of particular interest in our investigations, is the squared coefficient of variation (SCV) of the inter-departure process of a periodic probing stream. We analyze approximately the departure process of this probing stream. Simulation results indicate that the proposed hybrid approximation can provide good estimates of the SCV of the probing stream regardless of the stochastic behavior of the arrival process of the cross traffic. Given a measured SCV, inverting this approximation infers the load of the cross traffic on the congested link.

Towards unified self-congestion probing for bandwidth measurement

The self-congestion probing, which estimates bandwidth by controlling a temporal congestion of a probing stream, is the most popular approach in bandwidth measurement. Self-congestion tools are easy to implement, fast to converge and are robust to network dynamics with reasonably good accuracy; however, the current tools only exploit parts of the congestion signals, though the probing stream experiences a rich spectrum of congestion signals. TCP protocols follow the same self-congestion principle on inferring available bandwidth. We propose a unified self-congestion probing framework by bridging the self-congestion probing for available bandwidth measurement and TCP congestion control. The recent progress of TCP congestion control in both theory and experimentation provides new avenues in improving the current self-congestion tools. Based on this unified framework, we design and evaluate a simple available bandwidth probing scheme to utilize the explicit congestion notification signal, namely, ECNProbe. We conduct a measurement study on a Linux-based testbed and evaluate the performance of several available bandwidth measurement tools. We demonstrate that the proposed ECNProbe significantly improves the measurement accuracy with small convergence time and low overhead probing.

Earliest Deadline First Scheduling with Active Buffer Management for Real-Time Traffic in the Internet

We studied the problem of QoS guarantee for differentiated services. A two-level hierarchical scheduling framework was deployed to separate QoS metrics. Due to its desirable property of minimizing the maximum packet lateness, the Earliest Deadline First (EDF) scheduling was adopted to provide the in-class scheduling for the time-sensitive traffic. We employed an EDF scheduler combined with an active buffer management scheme (CHOKe) to improve the fairness of resource allocation and to maintain a good delay performance for real-time applications. Simulation results showed that the proposed scheme can achieve a better delay performance and make a more fair bandwidth allocation between the real-time TCP and UDP connections than the First Come First Served (FCFS) scheduling with the drop-tail buffer management which is commonly deployed in traditional IP routers.

The Earliest Deadline First Scheduling with Active Buffer Management for Real-Time Traffic in the Internet

We study the problem of QoS guarantee for differentiated services. A two-level hierarchical scheduling framework is deployed for the separation of the QoS metrics. Due to the desirable property of minimizing the maximum packet lateness, the Earliest Deadline First (EDF) scheduling is adopted to provide the in-class scheduling for the time-sensitive traffic. We propose to employ an EDF scheduler combined with an active buffer management scheme (CHOKe) to improve the fairness of resource allocation and to maintain a good delay performance for all real-time applications. Simulation results showthat the proposed scheme can achieve a better delay performance and make a more fair bandwidth allocation between the real-time TCP and UDP connections than the First Come First Serve (FCFS) scheduling with the Drop-Tail buffer management which is commonly deployed in the traditional IP router.