Distributed multi-source streaming models in peer-to-peer overlay networks
Introduction
Multimedia streaming applications [9], [12], [13] like music streaming and movie on demand are getting more significant than downloading service in the Internet applications from the security and financial reasons [1]. Here, multimedia contents have to be reliably delivered to users from providers of the contents while real-time constraints are satisfied. In peer-to-peer (P2P) overlay networks [2], [11], [16], multimedia contents are in nature distributed to peer computers (peers) in various ways like downloading and caching. Peers which have parts of multimedia contents can support other peers with the contents if it is legal. Here, a peer can play any role of contents and leaf. Contents peers support multimedia contents. Leaf peers receive multimedia contents from contents peers.
New approaches to realizing scalable and reliable multimedia streaming service in P2P overlay networks are discussed in multi-source streaming (MSS) models [5], [8] where multiple contents peers send packets of a multimedia content to a leaf peer. A large number of leaf peers are required to be supported by even low-performance, less reliable personal computers. Itaya et al. discuss a centralized protocol [5] similar to the two-phase commitment (2PC) protocol [14] to coordinate multiple contents peers. It takes at least three rounds to synchronize multiple contents peers. Liu and Voung [8] also discuss a centralized protocol where a requesting leaf peer sends a transmission schedule of a content to multiple contents peers. Each contents peer synchronously send packets according to the schedule. Although it is simple to implement the centralized approach, it takes time for multiple contents peers to start transmitting packets [3].
In the distributed multi-source streaming models [3], [4], [5], all the contents peers do not start transmitting packets of a multimedia content but contents peers gradually start transmitting packets. In this paper, we newly propose protocols based on gossip-based flooding protocols [6], [7] to reduce the communication overhead. First, a leaf peer sends a request packet to some number H of the contents peers. On receipt of the content request, a contents peer starts transmitting packets to the leaf peer. Here, each contents peer sends one Hth of the packets at one Hth of the transmission rate of the content rate, e.g. 30 Mbps of movie content. Then, the contents peer selects some number H of the contents peers. Here, there are two algorithms; a contents peer may be selected by multiple peers and by at most one peer. The former is a directed acyclic graph (DAG)-based coordination protocol (DCoP) and the latter is a tree-based coordination protocol (TCoP). A request packet carries information on which packets the contents peer has sent to a leaf peer at what rate. Each of the selected peers makes a decision on which packets to be sent at what rate by using the information. In addition, parity packets for some number of packets [15] are transmitted so that a leaf peer can receive every data of a content even if some number of packets are lost and contents peers are faulty. Furthermore, we discuss how to allocate packets of a multimedia content to each contents peer. Due to the heterogeneity of the P2P overlay networks, each contents peer sends packets at transmission rate and loss ratio different from others. We discuss a packet allocation algorithm where a leaf peer can deliver a packet on receipt of the packet without waiting for preceding packets from other contents peers even if some number of contents peers are faulty and packets are lost with a channel in a bursty manner.
In Section 2, we present the multi-source streaming (MSS) model. In Section 3, we discuss how to allocate packets of a content to contents peers in heterogeneous environment. In Section 4, we discuss a pair of the coordination protocols DCoP and TCoP. In Section 5, we evaluate the coordination protocols DCoP and TCoP in terms of how long it takes to synchronize all the contents peers and how many redundant packets are transmitted.
Section snippets
A multi-source streaming (MSS) model
Multimedia contents are distributed to peer computers (peers) in various ways like downloading and caching ones in a peer-to-peer (P2P) overlay network. For example, a peer obtains a free movie from an acquaintance peer by downloading and then supports some part of the movie to other peers [12]. A contents peer which holds whole or a part of a content can send the content to other peers. A peer receiving a content from a contents peer and using it is referred to as leaf peer. Each peer can play
Packet allocation to contents peer
Let pkt be a packet sequence 〈t1, …, tl 〉 (l ⩾ 1) of multimedia content c. Each contents peer CPi (i = 1, …, n) sends a part of a sequence pkt to a leaf peer LPs. Here, ∣pkt∣ = l. Suppose three contents peers CP1, CP2, and CP3 transmit packets in a sequence pkt = 〈t1, …, t7, …〉 of a multimedia content C to the leaf peer LPs where the bandwidth ratio bw1:bw2:bw3 is 4:2:1. Each contents peer CPi transmits a subsequence pkti of the packet sequence pkt to the leaf peer LPs. ∣pkti∣ ⩾ ∣pktj∣ if bwi ⩾ bwj. For example,
Types of distributed coordinations
Multiple contents peers CP1, …, CPn are required to cooperate to reliably deliver packets of a multimedia content C to a leaf peer LPs. Here, we assume each contents peer CPi supports the same transmission rate to the leaf peer LPs for simplicity. Let pkt be a sequence of packets t1, …, tl of a multimedia content C. While transmitting packets of the content C–LPs, each contents peer CPi informs the other contents peers of which packets CPi has sent at what rate and the view showing which contents
Evaluation
We evaluate a pair of the DAG-based coordination protocol (DCoP) and tree-based coordination protocol (TCoP) for synchronizing multiple contents peers in terms of the coordination time, the number of redundant parity packets, and receipt and transmission rates. Suppose there are n contents peers CP1, …, CPn which transmit packets of a content to a leaf peer LPs. Let H be the number of child contents peers to be selected by each parent (H ⩽ n). (H − h) shows packet interval. Suppose each communication
Concluding remarks
In this paper, we discussed the multi-source streaming model for transmitting continuous multimedia contents from multiple contents peers to a leaf peer. In peer-to-peer (P2P) overlay networks, peers on various types of computers can support other peers with multimedia contents. We discussed two types of the distributed coordination protocols, DAG-based coordination protocol (DCoP) and tree-based coordination protocol (TCoP) for multiple contents peers to transmit packets to a leaf peer. In
References (16)
- Apple Computer, Inc. iTunes,...
- I. Clarke, O. Sandberg, B. Wiley, T.W. Hong, Freenet: A distributed anonymous information storage and retrieval system,...
- S. Itaya, T. Enokido, M. Takizawa, A high-performance multimedia streaming model on multi-source streaming approach in...
- S. Itaya, T. Enokido, M. Takizawa, A. Yamada, A scalable multimedia streaming model based on multi-source streaming...
- S. Itaya, N. Hayashibara, T. Enokido, M. Takizawa, Scalable peer-to-peer multimedia streaming model in heterogeneous...
- et al.
Probabilistic reliable dissemination in large-scale systems
IEEE Trans. Parallel Distributed Systems
(2003) - M.-J. Lin and K. Marzullo, Directional Gossip: Gossip in a Wide Area Network, Technical Report: CS1999-0622,...
- X. Liu and S.T. Vuong, Supporting low-cost video-on-demand in heterogeneous peer-to-peer networks, in: Proceedings of...
Cited by (1)
GPM: A generic and scalable P2P model that optimizes tree depth for multicast communications
2012, International Journal of Communication Systems