01-09-2014, 03:35 PM
A DISTRIBUTED PROTOCOL TO SERVE DYNAMIC GROUPS
FOR PEER-TO-PEER STREAMING
A DISTRIBUTED.docx (Size: 82.53 KB / Downloads: 12)
ABSTRACT:
Peer-to-peer (P2P) streaming has been widely deployed over the Internet. A streaming system usually has multiple channels, and peers may form multiple groups for content distribution. In this paper, we propose a distributed overlay framework (called SMesh) for dynamic groups where users may frequently hop from one group to another while the total pool of users remain stable. SMesh first builds a relatively stable mesh consisting of all hosts for control messaging. The mesh supports dynamic host joining and leaving, and will guide the construction of delivery trees. Using the Delaunay Triangulation (DT) protocol as an example, we show how to construct an efficient mesh with low maintenance cost. We further study various tree construction mechanisms based on the mesh, including embedded, bypass, and intermediate trees. Through simulations on Internet-like topologies, we show that SMesh achieves low delay and low link stress.
INTRODUCTION:
With the penetration of broadband Internet access, there has been an increasing interest in media streaming services. Recently, P2P streaming has been proposed and developed to overcome the limitations of traditional server-based streaming. In a P2P streaming system, cooperative peers self-organize themselves into an overlay network via unicast connections. They cache and relay data for each other, thereby eliminating the need for resourceful servers from the system. Today, several practical P2P streaming software implementations have been shown to be able to serve up to thousands of peers with acceptable quality of service.
In a P2P streaming system, the server (or a set of servers) usually provides multiple channels. A peer can freely switch from one channel to another. For example, one of the most popular P2P streaming systems, PP Live, has provided over 400 channels according to a measurement study from the Polytechnic University, the total number of peers in PP Live during a day in 2007 varies from around 50 thousand to 400 thousand, and the number of peers in a single channel, e.g., CCTV1, varies from several hundred to several thousand. We can see that there is a large pool of peers in the streaming network. Peers are divided into multiple small groups, each corresponding to a channel. Peers in the same group share and relay the same streaming content for each other. In another study, a six-month 150- channel IPTV trace shows that people frequently change from one channel to another, with the median and mean channel holding time being 8 seconds and 14.8 minutes, respectively. In fact, there are many other similar applications over the Internet. In the application, the system contains multiple groups with different sources and contents. A user may join a specific group according to its interest. While the lifetime of users in the system is relatively long and the user pool is rather stable, users may hop from one group to another quite frequently.
In above applications, as peers may dynamically hop from one group to another, it becomes an important issue to efficiently deliver specific contents to peers. One obvious approach is to broadcast all contents to all hosts and let them select the contents. Clearly, this is not efficient in terms of bandwidth and end-to-end delay, especially for unpopular channels. Maintaining a separate and distinct delivery overlay for each channel appears to be another solution. However, this approach introduces high control overhead to maintain multiple dynamic overlays.
In this paper, we consider building a data delivery tree for each group. To reduce tree construction and maintenance costs, we build a single shared overlay mesh. The mesh is formed by all peers in the system and is, hence, independent of joining and leaving events in any group. This relatively stable mesh is used for control messaging and guiding the construction of overlay trees. With the help of the mesh, trees can be efficiently constructed with no need of loop detection and elimination. Since an overlay tree serves only a subset of peers in the network, we term this framework Subset-Mesh, or SMesh. Our framework may use any existing mesh-based overlay network. In this paper, we use Delaunay Triangulation (DT) as an example [10]. We propose several techniques to improve the DT mesh, e.g., for accurately estimating host locations and distributed partition detection. Based on the mesh, we study several tree construction mechanisms to trade off delay and network resource consumption.
We further present a distributed algorithm on how to detect and recover mesh partitions.
Our mesh has the following properties: - Low delivery delay: As mesh formation and message forwarding are based on hosts’ network locations, the delay for data delivery is significantly reduced as compared to that in the traditional DT mesh.
• Distributed: Unlike the traditional DT mesh, SMesh does not require a central server for mesh maintenance. It is fully distributed and scalable.
• Construction of data delivery trees: Given the mesh, we study how source-specific overlay trees can be efficiently constructed and maintained.
• We consider three ways to construct a tree:
1) Embedded tree, where tree branches are all mesh edges;
2) Bypass tree, where tree nodes can only be group members and tree branches may not be mesh edges; and
3) Intermediate tree, which is a trade-off between an embedded tree and a bypass tree.
These trees have the following properties:
Overhead reduction: As compared to traditional tree-based protocols, SMesh achieves much lower control overhead for tree construction and maintenance. This is because the mesh has maintained enough host information and can efficiently deal with host hopping between different groups.
QoS provisioning: SMesh provides QoS in the following senses:
1) It limits the node stress of a host in a tree according to the host’s capability.
2) It aggregates long-delay paths and delegates data delivery to shorter paths. As a result, packets may take more hops to reach their destinations, and this trades off end-to-end delay with network resource consumption. SMesh does not rely on a static mesh. In the case of host joining or leaving, the underlying DT mesh can automatically adjust itself to form a new mesh. The trees on top of it will then accordingly adjust tree nodes and tree edges. Also note that in SMesh a host may join as many groups as its local resource allows. If a host joins multiple groups, its operations in different groups are independent of each other.
EXISTING SYSTEM:
Peer-to-peer (P2P) networks have been adopted for Internet live video-streaming service, and several practical systems have been deployed in past years due to the inherent scalability and ease of deployment. However, most of these systems are commercial and proprietary, and hence little research was done in the area of characterizing practical system performance properties. In this article, we mainly present our experience on a practical P2P-based live video- streaming system called Grid Media, which was employed to broadcast live the Chinese Spring Festival Gala show over the Internet a trace study to understand the service capacity, quality of streaming service, connection heterogeneity, user geographic distribution, and request and online duration characteristics. Our observations shed light on those systems and further improvements in the arena of large-scale live video-streaming service over the Internet
CONCLUSION:
In P2P streaming networks, users may frequently hop from one group to another. In this paper, we propose a novel framework called SMesh to serve dynamic groups for Internet streaming. SMesh supports multiple groups and can efficiently distribute data to these dynamic groups. It first builds a shared overlay mesh for all hosts in the system. The stable mesh is then used to guide the construction of data delivery trees for each group. We study three ways to construct a tree, i.e., embedded, bypass, and intermediate trees. We also propose and study an aggregation and delegation algorithm to balance the load among hosts, which trades off end-to-end delay with lower network resource usage. Through simulations on Internet-like topologies, we show that SMesh achieves low RDP and low link stress as compared to traditional tree-based protocols.