An adaptive peer-to-peer live streaming system with incentives for resilience

Abstract

Recently, there have been a lot of research efforts on peer-to-peer (P2P) live streaming services. P2P systems can be easily deployed since a participating peer’s resources (i.e., upload link bandwidth) can be exploited to distribute contents. However, how to adapt to leaving peers and how to encourage peers to contribute resources voluntarily are still challenging issues. In this paper, we propose Climber, an adaptive P2P live streaming system with incentives for resilience. Climber is based on the hybrid structure of a tree and a mesh, so as to achieve self-improvement and adaptation to users’ dynamic joining and leaving. Moreover, Climber substantiates an incentive mechanism that provides better resilience for peers with more upload bandwidth allocated. Simulation results reveal that Climber significantly reduces the topology maintenance cost compared to SplitStream and NICE-PRM. Also, simulation and analytical results verify that Climber can bound the level of disruption by dynamically adapting to the user churning rate.

Introduction

Recently, live streaming services over the Internet (e.g., sports game or music concert) are becoming popular. For example, Akamai [2] provides an infrastructure support for these services; AOL broadcast [3] and MSN broadcast [4] serve as streaming portals. Live Earth concerts in July 2007 on MSN broadcast service recorded 15 million streaming sessions, with 237,000 simultaneous viewers at the peak [4]. These services leverage infrastructures (i.e., server-farm based solutions or content delivery networks (CDNs)) where contents are replicated on a number of servers to improve the speed of content delivery. However, infrastructure-based live streaming systems have some drawbacks in terms of cost, scalability, and deployment. Akamai reports an aggregate traffic of 885 Gbps during peak traffic periods through 30,000 servers in 67 countries [5].

Unlike these infrastructure-based services, peer-to-peer (P2P) live streaming is gaining much attention in the literature [6], [7], [8], [9], [10] because of its scalability, low cost, and tactical deployment. In P2P live streaming, all peers participate in distributing contents by sharing their resources (i.e., upload link bandwidth). However, providing a resilient P2P streaming service is a key challenge since peers are prone to departures and failures. Especially, a peer acts as both a server and a client, and thus each failure of a peer causes streaming disruption for peers downstream until the time the relevant topology is reconstructed.

In this paper, we focus on two challenging issues for resilient P2P live streaming: (1) efficiency and (2) incentive. The first issue is how to efficiently support dynamic membership; i.e., each peer dynamically joins or leaves (peers easily fail or leave at will) the P2P network. Hence resilience to churning rate should be taken into account. In general, sending more signaling packets (e.g., keep-alive or heartbeat messages) between peers will increase the resilience level of the network, while consuming a considerable amount of network resources. There is a trade-off between the service resilience and the signaling overhead of the system; so the target level of resilience should be carefully decided. The control overhead of a system is also tightly coupled with the complexity of the system. The system should be designed to be as simple as possible for efficient operations.

The second issue, incentive, is more crucial. The performance of P2P systems strongly relies on how much resources are contributed by peers. Therefore, the way to encourage peers to contribute their resources for stream distribution should be devised in order to improve the quality of live streaming services in terms of delay and resilience. In P2P file-sharing applications (e.g. [11], [12]) with an incentive mechanism, download rate of a peer is roughly proportional to the peer’s contributions. Hence, peers with more contributions need a shorter time to receive a file than others. Even though a peer with low contribution (i.e., a peer with small upload capacity) is able to receive a complete file, it will take longer time than peers with high contributions. This incentive mechanism makes sense since the objective is just to share the file. However, there is a significant difference in live streaming applications and file-sharing applications. Especially in the P2P live streaming applications, the incentive mechanism should be designed by considering unique characteristics of P2P live streaming which contrast with those file-sharing applications.

There have been some P2P live streaming proposals that have similar incentive mechanisms to the file-sharing applications. Generally, peers contribute to a system by providing their upload bandwidth in order to gain reputation (e.g. [13], [14]) or currency (e.g. [15]) as a reward. In these systems, reputation is used to compete for connections with better neighbors against other peers, or currency is consumed to buy chunks from other peers. However, streaming packets of a stream are generated by a source in live streaming applications, encoded and sent to other peers (suppose the stream is k kbps). Live streaming packets are periodically generated and streaming packets at the current moment are disseminated. Thus, if a peer with high contributions is receiving a stream of k kbps, there is no way to give more incentive to this contributor. Peers with low contributions, receiving lower than k kbps, cannot play a video due to the nature of live streaming. Therefore, we should devise a different incentive mechanism for live streaming applications.

We propose a novel P2P live streaming system, Climber. The salient characteristics of Climber are summarized as follows. (1) Climber is a hybrid of a tree and a mesh. That is, a single tree is a main structure for streaming; however, random (or redundant) edges between peers (not between parent and child) are added to the tree for resilient connectivity. (2) Climber is simple in the sense that the effect of the dynamic membership on the system topology is mitigated. Climber simply attaches a new joining node¹ to the tree as a leaf node. When a non-leaf node leaves or fails, Climber minimizes the effect of peer departure by efficiently relocating its descendants to other positions of the tree. (3) Despite the time-varying network conditions (e.g., churning rate, congestion, delay), Climber bounds the level of resilience of a system. Climber adaptively changes its mesh topology (i.e., the random edges) to maintain a target level of resilience which is an input QoS parameter. (4) Climber keeps improving the transfer delay of a tree using the mesh topology. When a peer finds an alternative path that is faster than the current data path from the root, the peer switches its link to the alternative path. This switching process continuously improves the overall root-to-peer (or end-to-end) delay in the system. (5) Climber embodies an incentive mechanism that enhances resilience despite churning rate. The number of potential incoming data paths from a source to a peer is approximately proportional to the number of child nodes of the peer. Hence, a peer will experience more resilient services as the peer maintains more child nodes (or allocate more upload link bandwidth). This incentive mechanism encourages nodes to assign more upload link bandwidth for streaming data distribution, which leads to more (system-wide) resilient services.

The contribution of this paper is three-fold. First, we proposes a simple and adaptive system for resilient P2P streaming. Second, an analytical model, which demonstrates the effect of the disruption rate, is developed and validated by simulations. Third, extensive simulation results are given to evaluate the performance of Climber against other schemes. To the best of our knowledge, Climber is the first proposal that provides the required level of resilience probabilistically, and the incentives to contribute more resources for better resilience that suits P2P live streaming services.

The rest of this paper is organized as follows. Design concepts and the operations of Climber are introduced in Section 2, and we analyze the level of disruption of Climber in Section 3. Section 4 gives simulation results, followed by related work in Section 5. Finally, we conclude this paper in Section 6.