Bisimulations for Communicating Transactions

Abstract

Abstract. We develop a theory of bisimulations for a simple language containing communicating transactions, obtained by dropping the isolation requirement of standard transactions. Such constructs have emerged as a useful programming abstraction for distributed systems. In systems with communicating transactions actions are tentative, waiting for certain transactions to commit before they become permanent. Our theory captures this by making bisimulations history-dependent, in that actions performed by transactions need to be recorded. The main requirement on bisimulations is the systems being compared need to match up exactly in the permanent actions but only those. The resulting theory is fully abstract with respect to a natural contextual equivalence and, as we show in examples, provides an e↵ective verification technique for comparing systems with communicating transactions.

Reversible client/server interactions

“…In [LMSS11] Lanese et al enrich the calculus of [LMS10] with a fine-grained rollback primitive. To the best of our knowledge the first works dealing with rollback of communicating systems are [dVKH10a,dVKH10b,KSH14]. In these papers an extension of CCS models the combination of rollback recovery and coordinated checkpoints.…”

Distinguishing between communicating transactions

“…Transactional constructs without the isolation principle have been proposed as useful building blocks of distributed systems (e.g., [10,11,17,23,3,6]). Communicating transactions is such a construct, equipped with a rich theory providing techniques for proving behavioural equivalence of transactional systems [8,9,16]. To develop useful verification tools, however, it is also essential to have techniques for exhibiting the in-equivalence of systems, rather than relying on the absence of equivalence proofs.…”

“…However if this c-action is performed then the transaction can never commit (i.e., perform a co-action) and therefore the presence of this potential c-action is superfluous. According to TCCS m reduction barbed equivalence theory [16] these two transactions are behaviourally equivalent. Consequently an extension of HML we propose should not be able to distinguish them, despite the fact that Q 1 can apparently perform a c-action or at least attempt to do so.…”

“…The notion of weak bisimulation developed in [16] for transactions contains constraints on the actions which transactions may perform-the standard transfer property. This in effect compares the future behaviour of transactions.…”

“…In the latter configuration, the two transactions have been merged and thus obtained the same name l. After a single commit, C 1 becomes a, l 2 (b) • 0 | co l2 0 where only past action a is committed. Because there is no matching future configuration of C 2 , weak bisimulation from [16] distinguishes the two processes. Thus, to distinguish P 2 from Q 2 , one would expect a property logic for transactions containing, in addition to standard future-oriented modal operators discussed above, operators for examining past behaviour.…”