Article Preview
TopIntroduction
The literature review and analysis of the field of transactional systems and pervasive environment issues evince that transactional models and related techniques have constantly been confronted with the versatile application needs in terms of transactional properties (Karlsen, 2003; Santos, Veiga, & Ferreira, 2004; Rouvoy, Serrano-Alvarado, & Merle, 2006). Indeed, with the advent of mobile computing, the dynamic nature of application execution context induces multiple transactional needs during the lifecycle of a transaction. Transactions are powerful abstractions that offer guarantees in terms of reliability to database systems and distributed systems in the presence of concurrency and faults. These guarantees are obtained by the implementation of ACID properties (Atomicity, Coherence, Isolation, Durability) (Gray & Reuter, 1993). The atomicity property of a distributed transaction can only be assured by an atomic commit protocol. The latter has a considerable impact on the performance of distributed systems. Indeed, the theoretical and practical aspects of this problem, coupled with recent developments in pervasive environments have renewed the motivation for research on this issue.
As in other distributed protocols, the classical 2PC protocol (Liu, Agrawal, & El Abbadi, 1998) is based on the assumption that all parties involved in the implementation of the protocol are equipped with sufficient computing and power resources, and messages are exchanged over wired networks with available bandwidth permanently (Chahbour & Nouali, 2003). These assumptions are no longer verified in pervasive environments where the sites may be mobile units, weakly equipped with resources (CPU, memory, and energy) and communicate via wireless links. This wireless communication leads to lower bandwidth, latency, higher error rates and higher cost. Furthermore, to meet the variables requirements of transactional services, the need to relax ACID properties has been proposed in many researches since the early 90s. There was a great effort on extended transaction models (Elmagarmid, 1992; Chrysanthis & Ramamrithan, 1994). This effort has been continued more recently in the context of mobile computing to satisfy the constraints of the execution environment (Segun, Hurson, & Spink, 2001). Researches have led to different notions of atomicity (strict, relaxed and semantic atomicity), consistency (strict or weak), and isolation (strict or relaxed allowing a flexible interleaving between transactions and a controlled sharing of intermediate results) (Younas, Chao, Lo, & Li, 2006). Several standards specifications have been proposed, including WS-Transaction specification (Sun, 2006) and Business Transaction Protocol (Oasis Committee Specification, 2002). However, current standards don’t consider the context information. Several transactional models and techniques have been proposed (Schafer, Dolog, & Nejdl, 2008; Lakhal, Kobayashi, & Toyota, 2009), but they have limitations, namely, a non-consideration of the context information and the conception of advanced models with transactional properties that differ from one application to another.