chun Li

LPDHT: A Locality-Aware and Partitioned-Space Architecture for Peer-to-Peer SIP

SIP is a signaling protocol widely used in multimedia communication. Recently, P2PSIP, which combines DHT and SIP, has been proposed to overcome the drawbacks of traditional CS SIP. Although P2PSIP has many advantages, the introduction of P2P also brings in some disadvantages in performance. And these disadvantages become unbearable in a huge P2PSIP overlay in Internet environment. To address these problems, we propose a novel DHT architecture for P2PSIP, LPDHT, whose full name is Locality-Aware and Partition-Space DHT. In the super nodes form P2PSIP, LPDHT can reduce not only DHT routing hop count and latency per DHT hop, but also the overall overhead. Therefore it can improve system capacity greatly. Performance analysis indicates that overall overhead are reduced dramatically in LPDHT.

Architecture Design of P2PSIP System

P2PSIP (Peer-to-Peer Session Initiation Protocol) is proposed to construct fully distributed signaling architecture for multimedia communication systems. Instead of centralized SIP servers, P2PSIP overlays utilize distributed and dynamic nodes to support location and storage services. So far various design alternatives of overlay architectures have been proposed to describe the mechanism about how the user nodes are connected together to form an application layer topology. However, there appears to have been little research work on comprehensively comparing the performance properties of various P2PSIP architectures, and no study has been made to investigate how to select appropriate overlay architecture according to application requirements. In this paper, we investigate various P2PSIP architectures and compare their features mainly from the perspective of application performance. The query latency of overlay and call setup delay of SIP sessions are modeled as significant performance considerations for selection of suitable architecture for different usage scenarios. First, we comprehensively analyze implementations of P2PSIP architecture. Four typical P2PSIP architectures are described in detail: pure DHT, DHT-based SIP server farm, P2PSIP overlay formed by super nodes, and hierarchical P2PSIP overlay. Furthermore, we classify the nodes into peer and client according to their roles in the architecture. Then the performance model of query latency and call setup delay for P2PSIP overlay architectures is proposed. Several parameters to present features of the overlay architectures are defined. The analysis primarily focuses on the impact of possibility of packet loss caused by link error or network congestion and churn of overlay. Finally, an event-driven chord-based simulation to evaluate query latency and call setup delay for different P2PSIP architectures is developed. An analytical model and simulation showed that the average call setup delay is less than ten seconds for all the architectures, which could satisfy the requirements of multimedia session establishments in most cases. We advocate that the super node or hierarchical architecture might be appropriate options rather than pure DHT in terms of latency performance in heterogeneous network environments. The server farm approach is good at management as well as performance, which might be the best option for systems supported by service providers.

Effective Data Replication in Heterogeneous Structured P2P Networks

The traditional way to achieve high data availability in structured P2P network is to replicate the data items preserved on the node to its neighbor nodes. However, in some heterogeneous structured P2P networks, the neighbor nodes are always close to each other in locality. Therefore, if some local disaster happened, it is very likely that some data items would be permanent lost. Moreover, this traditional data replication strategy will bring complex data migration when there are nodes join or leave the DHT network. In this paper, we propose a novel data replication strategy in heterogeneous structured P2P networks. It not only guarantee the availability of data items in the situation of large-scale local disaster, but also greatly reduce the complexity of data migration when nodes join or leave the network.