Concurrent models for Linda with transactions

摘要

Nowadays concepts, languages and models for coordination cannot leave aside the needs of the increasing number of commercial applications based on global computing over the (inler)net. For example, platforms like Microsoft .NET and Sun Microsystems Java come equipped with packages for supporting ad hoc transactional features, which are essential for most business applications. We show how to extend the coordination language par excellence (viz. Linda) with basic primitives for transactions, while retaining a formal model for its concurrent computations. This is achieved by exploiting a variation of Petri nets, called zero-safe nets, where transactions can be suitably modelled by distinguishing between stable places (ordinary ones) and zero places (where tokens can only be temporarily allocated, defining hidden states). The relevance of the transaction mechanism is illustrated in terms of expressive power. Finally, it is shown that stable places and transactions viewed as atomic steps define an abstract semantics that is apt for a fully algebraic treatment, as demonstrated oia categorical adjunctions between suitable categories of nets.