Lisong Xu

All-Optical Network Coding

We investigate the application of network coding to all-optical networks from both the algorithmic and infrastructural perspectives. We study the effectiveness of using network coding for optical-layer dedicated protection of multicast traffic that provides robustness against link failures in the network. We present a heuristic for solving this problem and compare it with both inefficient optimal methods and non-network-coding approaches. Our experiments show that our heuristic provides near-optimal performance while significantly outperforming existing approaches for dedicated multicast protection. We also propose architectures for specialized all-optical circuits capable of performing the processing required for network coding and show how these devices can be effectively deployed in an all-optical multicast network.

TCP Congestion Avoidance Algorithm Identification

The Internet has recently been evolving from homogeneous congestion control to heterogeneous congestion control. Several years ago, Internet traffic was mainly controlled by the traditional AIMD algorithm, whereas Internet traffic is now controlled by many different TCP algorithms, such as AIMD, BIC, CUBIC, and CTCP. However, there is very little work on the performance and stability study of the Internet with heterogeneous congestion control. One fundamental reason is the lack of the deployment information of different TCP algorithms. In this paper, we first propose a tool called TCP Congestion Avoidance Algorithm Identification (CAAI) for actively identifying the TCP algorithm of a remote web server. CAAI can identify all default TCP algorithms (i.e., AIMD, BIC, CUBIC, and CTCP) and most non-default TCP algorithms of major operating system families. We then present, for the first time, the CAAI measurement result of the 5000 most popular web servers. Among the web servers with valid traces, we found that only 16.85~25.58% of web servers still use the traditional AIMD, 44.51% of web servers use BIC or CUBIC, and 10.27

Linear Programming Models For Multi-Channel P2P Streaming Systems

sim

A flexible Divide-and-Conquer protocol for multi-view peer-to-peer live streaming

19% of web servers use CTCP. In addition, we found that, for the first time, some web servers use non-default TCP algorithms, some web servers use some unknown TCP algorithms which are not available in any major operating system family, and some web servers use abnormal slow start algorithms. Our CAAI measurement results show a strong sign that the majority of TCP flows are not controlled by AIMD anymore, and a strong sign that the Internet congestion control has already changed from homogeneous to highly heterogeneous.

Network coding for optical-layer multicast

Most of the commercial P2P video streaming deployments support hundreds of channels and are referred to as multichannel systems. Measurement studies show that bandwidth resources of different channels are highly unbalanced and thus recent research studies have proposed various protocols to improve the streaming qualities for all channels by enabling cross-channel cooperation among multiple channels. However, there is no general framework for comparing existing and potential designs for multi-channel P2P systems. The goal of this paper is to establish tractable models for answering the fundamental question in multi-channel system designs: Under what circumstances, should a particular design be used to achieve the desired streaming quality with the lowest implementation complexity? To achieve this goal, we first classify existing and potential designs into three categories, namely Naive Bandwidth allocation Approach (NBA), Passive Channel-aware bandwidth allocation Approach (PCA) and Active Channel-aware bandwidth allocation Approach (ACA). Then, we define the bandwidth satisfaction ratio as a performance metric to develop linear programming models for the three designs. The proposed models are independent of implementations and can be efficiently solved due to the linear property, which provides a way of numerically exploring the design space of multi-channel systems and developing closedform solutions for special systems.

Variable neighbor selection in live peer-to-peer multimedia streaming networks

Multi-view peer-to-peer (P2P) live streaming systems have recently emerged, where a user can simultaneously watch multiple channels. Previous work on multi-view P2P streaming solves the fundamental inter-channel bandwidth competition problem at the individual peer level, and thus can be used with very limited types of streaming protocols. In this paper, we propose a new protocol for multi-view P2P streaming, called Divide-and-Conquer (DAC), which efficiently solves the inter-channel bandwidth competition problem using a divide-andconquer strategy at the channel level, and thus is flexible to work with various streaming protocols. This makes DAC more suitable for upgrading current single-view P2P live streaming systems to multi-view P2P live streaming systems. Our extensive packetlevel simulations show that DAC is efficient in allocating the overall system bandwidth among competing channels, is flexible in working with various streaming protocols, and is scalable in supporting a large number of users and channels.

Channel-Aware Peer Selection in Multi-View Peer-to-Peer Multimedia Streaming

The network coding paradigm has become an effective method for achieving efficient multicast in communication networks. The optical community has just started to venture into the application of network coding in optical networks. However, a number of challenges need to be overcome before network coding can be used in optical networks. These include limited buffering and processing capabilities as well as extremely coarse bandwidth granularity. In this paper, we address some of these problems. Finding multicast codes can be broken into two subproblems: finding a subgraph of the topology to code over and then finding an actual code for that subgraph. We show that the former problem is NP-Complete and provide heuristics which allow for coded multicast in optical wavelength division multiplexing networks which offer a modest improvement in bandwidth efficiency over traditional methods for finding routes for optical-layer multicast traffic.

Improving multi-view peer-to-peer live streaming systems with the divide-and-conquer strategy

Data-driven (or swarming based) streaming is one of the popular ways to distribute live multimedia streaming traffic over peer-to-peer (P2P) networks. The efficiency and user satisfaction highly depend on the constructed overlays. The common neighbor selection algorithms in existing overlay construction schemes usually randomly select a fixed number of neighbors which satisfy the selection requirements, such as end-to-end delay or a peerpsilas sojourn time. However, this fixed random neighbor-selection algorithm (FRNS) neglects the peerspsila upload bandwidth heterogeneity and therefore, the upload bandwidth cannot be efficiently used. In this paper, we propose a variable random neighbor-selection (VRNS) scheme to alleviate the problems due to bandwidth heterogeneity, and in which the number of neighbors with different upload bandwidths is dynamically determined by the statistical bandwidth information of the system. Our proposed scheme is shown to outperform FRNS based upon a large volume of carefully designed simulations.

Comparing multi-channel Peer-to-Peer video streaming system designs

Motivated by the success of the Picture in Picture feature of the traditional TV, several commercial Peer-to-Peer MultiMedia Streaming (P2PMMS) applications now support the multi-view feature, with which a user can simultaneously watch multiple channels on its screen. This paper considers the peer selection problem in multi-view P2PMMS. This problem has been well studied in the traditional single-view P2PMMS; however, it becomes more complicated in multi-view P2PMMS, mainly due to the fact that a peer watching multiple channels joins multiple corresponding overlays. In this paper, we propose a novel peer selection algorithm, called Channel-Aware Peer Selection (CAPS), where a peer selects its neighboring peers based on the channel subscription of the system, in order to efficiently utilize the bandwidth of all peers in the system, especially those peers watching multiple channels. The results of a large-scale simulation with 10,000 peers and 4 channels shows that CAPS can significantly improve the system performance over the straightforward Random Peer Selection (RPS), which is widely used in single-view P2PMMS networks.

Providing statistically guaranteed streaming quality for peer-to-peer live streaming

Multi-view peer-to-peer (P2P) live streaming systems have recently emerged, where a user is allowed to simultaneously watch one or multiple channels. Previous work on building multi-view P2P streaming systems requires network coding in data block scheduling within each channel, since it solves the fundamental inter-channel bandwidth competition problem at the individual peer level. Therefore, it limits the migration from existing single-view systems to multi-view systems. In this paper, we propose a new protocol, called divide-and-conquer (DAC), which provides a flexible way of building multi-view P2P streaming systems based on existing single-view systems. Specifically, the DAC protocol solves the inter-channel bandwidth competition problem at the channel level based on a divide-and-conquer strategy. Our extensive packet level simulations show that our DAC protocol achieves the three design goals: flexibility, efficiency and scalability.