Transactional support for cooperative applications

The rapid growth of the global information infrastructure opens new opportunities to perform joint work in areas such as cooperative authoring, distributed software development, and business workflows. Cooperative work on shared data requires computer support to coordinate the work of multiple users and to ensure the consistency of data processed within a cooperative application. Maintaining data consistency in multi-user systems is the classical problem of transaction management. The traditional transaction model addresses environments in which a large number of relatively short-lived transactions access a shared database. It is based on the fundamental properties of atomicity, consistency, isolation, and durability of transactions. However, this transaction model does not meet the requirements of cooperative applications. Isolation of transactions, as guaranteed by the ACID properties, contradicts the need to cooperate. Atomicity of transactions is too strict for interactive environments, where activities are of long duration. A cooperative transaction model needs to support the interactive execution of longrunning activities in which competition for resources, on which the locking protocols are based, is replaced by the need to cooperate. The emphasis, therefore, is not on preventing access to resources, but rather on the semantically correct exchange of information among concurrent activities of cooperating users. In this dissertation, we present a semantics-based cooperative transaction model, called COACT, that addresses the key features of cooperative applications, i.e., interactive user control, long-duration activities, and multiuser cooperation on shared persistent data. The concepts presented in this thesis are preceded by a thorough analysis of requirements derived from a broad spectrum of cooperative applications, ranging from cooperative hypermedia authoring to workflow applications. The COACT cooperative transaction model builds on the observation that cooperative work is characterized by alternating periods of individual and joint work. Therefore, COACT combines a workspace concept with a new synchronization mechanism, called history merging. To allow for individual work, we assign a private workspace to every user participating in a cooperative activity. Additionally, there is a common workspace accessible by all participants. To enable joint work, COACT provides a set of information exchange primitives that are all based on history merging. These primitives allow group members to exchange information among workspaces in order to combine their work into a coherent whole. To that end, we take an operation-oriented view in COACT. State transitions on workspaces are modeled as operations and are recorded in workspace histories. Information exchange among workspaces is realized by the exchange of operations among workspaces, instead of data objects. The consistency of common work results is determined on the basis of the semantics of operations performed in the workspaces. COACT utilizes both backward and forward commutativity relations for this. The flexibility of our approach is mainly achieved by its consideration of operations semantics for dynamically determining consistent units of work, for detecting and resolving conflicts during a merge, and for selective compensation of operations in workspaces. Merging histories with pre-specified commutativity relations provides transactional guarantees to a cooperative activity. In particular, we guarantee in COACT that all workspace histories remain legal. Using legality as correctness criterion, history merging relaxes the isolation and atomicity properties of the traditional transaction model, while ensuring consistency of the workspaces. History merging also provides a basis for management of consistency between disconnected or mobile users who operate independently and, yet, must occasionally reconcile their work with each other or with the stationary system. We show in this thesis how to adapt the COACT cooperative transaction model to the special characteristics of mobile environments in order to support disconnected operation mode. Besides the conceptual development of a formal transaction model, a proof-of-concept demonstrator has been implemented. We describe in this dissertation how the COACT cooperative transaction model has been implemented using an object-oriented database management system and how the new cooperation facilities can be integrated into a hypermedia authoring system. We believe that the integration of a cooperative transaction model like COACT with advanced database technology can ease the development and deployment of cooperative applications. Furthermore, history merging provides directions for solving open research problems, like semi-automatic merging of versions and re-integration support in mobile environments and asynchronous replication