W. Janssen, M. Poel and J. Zwiers. Action Systems and Action Refinement in the Development of Parallel Systems. Proceedings of CONCUR 1991, LNCS 527, 298--316, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....also. The point is however that there is nothing optional about the refinement operator: it will always yield the same system, regardless whether this is what the designer had in mind or not. We should mention a paper formalizing a notion of abstraction much like ours: Janssen, Poel and Zwiers [10]. Their approach is based on a concrete interpretation of the notion of causality: one event must be causally related to another if and only if they share so called resources. An event transformation is valid if and only if the high level event and its lowlevel behaviour have exactly the same ....

W. Janssen, M. Poel, and J. Zwiers. Actions systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298--316. Springer-Verlag, 1991.

....set of traces, constituting the possible runs of the system. Thus, traces are essentially a linear time model for behaviour: the moments of choice in a behaviour are not represented. A process algebraic, linear time setting with action dependencies has been developed by Janssen, Poel and Zwiers in [47, 48]. In particular, they also introduce an operator for action refinement which takes action dependencies into account. In the present paper, we develop a branching time semantics for a process algebra based on action dependencies, containing most of the standard features, such as sequential ....

....overlap, controlled by the appropriate action dependencies. Then, we show that this principle can be applied more generally, by formulating a version of the communications closed layers law proposed by Elrad and Francez [31] and promoted (in a linear time setting) by Zwiers et al. for instance in [47]. Finally, we also give an example from the field of data bases, showing the decomposition of an atomic query. 2 Language We start with the introduction of our process algebra and give a first informal discussion of its semantics. Act denotes a set of actions, ranged over by a, b, c, The ....

[Article contains additional citation context not shown here]

W. Janssen, M. Poel, and J. Zwiers. Actions systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298-316. Springer-Verlag, 1991.

....of refined actions is different from the serial order. This corresponds to the notion of serializable schedules in concurrency control theory. In process theory it has been introduced as a so called loose sequential operator. The original work about loose sequential operators in process algebra is [29]. This work has been further developed in [1] and [38] 38] generalizes [29] for branching time models. Unfortunately, the cited studies do not consider data refinement. Abstract data type refinement has already been thoroughly investigated for many years. Starting with Hoare s abstraction ....

....notion of serializable schedules in concurrency control theory. In process theory it has been introduced as a so called loose sequential operator. The original work about loose sequential operators in process algebra is [29] This work has been further developed in [1] and [38] 38] generalizes [29] for branching time models. Unfortunately, the cited studies do not consider data refinement. Abstract data type refinement has already been thoroughly investigated for many years. Starting with Hoare s abstraction function [24] several papers followed which examine the refinement of algebraic ....

W. Janssen, M. Poel, and J. Zwiers. Action Systems and Action Refinement in the Development of Parallel Systems. In J.C.M. Baeten and J.F. Groote, editor, CONCUR'91, pages 298--316. Springer, 1991. LNCS 527.

....of a single action or operation by a (partially ordered) set of actions. This would limit the resulting trace to orderings as in (3a) whereas our trace transformation can also result in orderings such as in (3b) 3c) and (3d) A more powerful notion of action refinement is introduced in [5] which is more like our trace transformation, in the sense that in addition to the substitution of a single action there is also a transformation of the dependencies. In the ESPRIT OMI COSY project applications are specified using the same model as in SPADE. They have also defined a refinement of ....

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems; an algebraic approach. In J. Baeten and J. Groote, editors, Proc. 2nd Int. Conf. on Concurrency Theory (CONCUR'91), LNCS 527, pages 298--316, Amsterdam, The Netherlands, Aug. 1991.

.... instance, if S = a; b and r(a) a 1 ; a 2 as above, then the only possible implementation is r(S) a 1 ; a 2 ; b and not the alternative a 1 ; a 2 b) where the causal relation between a and b is partially forgotten (see above) This can be a serious drawback in practice, as pointed out in [28]. Some work has been devoted to define less rigid forms of action refinement that do allow some overlapping [13, 27, 34, 39, 44] Still, 4 RENSINK AND GORRIERI in all these approaches, given a specification and a refinement function there is always only one possible implementation. By allowing ....

....of (the refinement of) the first operand and (the implementation of) the second operand. In other words, this rule expresses a certain degree of weakening of the causal ordering during refinement in the sense discussed in the introduction, reminiscent of the approach of Janssen, Poel and Zwiers [28] and Wehrheim [44] Note that, because of our choice of refinement language R, it is clear that u 1 is not terminated in the condition r(a) u 1 ; u 2 . This is indeed a necessary circumstance. To see why, note that we do also not expect the following generalisation of R 12 to VERTICAL ....

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298--316. Springer-Verlag, 1991.

....preserving bisimulation is known to be invariant against refinement and in [5] it is shown that this also holds for causal testing. However, one application of global dependencies is the definition of more loose notions of refinement (inheriting causalities only when specified by a dependency [7, 19]) In [20] invariance against loose refinement is shown to hold for history preserving bisimulation on prime event structures, for other models like e.g. flow event structures it is however lost. Another consequence of our result for bisimulation is that we can use Hennesssy Milner logic [6] as a ....

Wil Janssen, Mannes Poel, and Job Zwiers. Actions systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298--316. Springer-Verlag, 1991. 6

....Thus we call this type of refinement parametric. In contrast to standard action refinement parametric refinement is non strict and hence permits the top down design of systems in which independent actions will indeed be unordered. Our approach is a generalization of the approach by Zwiers et al. [13], 24] in the sense that our model reflects the branching structure of processes, whereas the model of Zwiers does not represent the points of choice between alternative behaviours. Besides the derivation of an implementation from a given specification, it is also important to be able to verify ....

....fa; bg is a process that tried to perform a and b but deadlocked (an equivalent process would be (a Delta b) k fa;bg (b Delta a) where a and b are independent) is the CCS choice operator. The operator Delta denotes weak (or parametric) sequential composition (called conflict composition in [13]) In the process P Delta Q the actions of P and Q are only causally dependent if they are D related. So it may be possible for that process to perform some Q actions before P has terminated and in fact sometimes even before P has started. The standard sequential composition can be obtained by ....

[Article contains additional citation context not shown here]

Wil Janssen, Mannes Poel, and Job Zwiers. Actions systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298--316. Springer-Verlag, 1991.

....when dealing with distributed systems, which include the nonsequential concept of parallelism. An algebraic approach to the specification and verification of distributed systems is the use of process algebraic languages, i.e. languages comprising combinators to build processes from other processes [1, 5, 11, 17, 18]. This paper investigates the usage of relational algebra as a process algebraic language. Relational algebra has a long mathematical tradition. Its present form is established by Tarski inspired by considerations of Schroder. The structure of relation algebras has been thoroughly studied in the ....

.... proposition P sat S, where P stands for a process term and S stands for the prescribed process semantics, as P ae S, where ae is a presupposed ordering on the considered model expressing the ordering of observations, which is the view as expected from implementing refinement propositions (cf. [17, 18, 11]) The starting point of our investigation is the comparison of the two signatures of ordinary process algebra (e.g. PA [1] and relational algebra (notations adopted from [25] It yields the following results: on the one hand, for relational algebra the parallel merge operator is missing, on the ....

[Article contains additional citation context not shown here]

Janssen, W., Poel, M., Zwiers, J., Action systems and action refinement in the development of parallel systems, in: Baeten, J.C.M, Groote, J.F. (eds.), CONCUR '91, Lecture Notes in Computer Science 527 (1991) 298--316, Springer-Verlag

.... . MP S = Causal(M S , S) M S . S. Union composition, expresses the program nondeterminism, concurrency: M S = Concurrent(M S1 , M S2 ) M S1 [ M S2 . A Meta Program, is rewritten MP S = Causal(M S , S) Causal(Concurrent(M S1 , M S2 ) Concurrent(S1, S2) The algebraic laws of CCL [3], extended in DUALITY, allow transformations of the Meta Program in forms that match shared memory or distributed parallel architectures. We obtain the DUALITY model in practice: MP S = M S . S (M S1 [ M S2 ) S1 [ S2) M S1 . S1) M S2 . S2) MP 1 [ MP 2 . Consistent coordination ....

Janssen, W., Poel, M, Zwiers, J., "Actions system and action refinement in the development of parallel systems", In Proc. of CONCUR'91, 298-316, LNCS 527, 1991.

.... among specification or mapping components is represented with union composition, MS = Concurrent(M S1 , M S2 ) M S1 [ M S2 ) We re write a Meta Program: MPS = Causal(MS , S) Causal(Concurrent(M S1 , M S2 ) Concurrent(S 1 , S 2 ) We extend in UNITY the algebraic laws of CCL [4] to allow transformations of the Meta Program in forms that match mobile distributed target architectures. Thus, we obtain the DUALITY model in practice: MPS = MS . S = M S1 [ M S2 ) S 1 [ S 2 ) M S1 . S1) M S2 . S2) MP 1 [ MP 2 . Consistent coordination: if the architecture ....

Janssen, W., Poel, M, Zwiers, J., "Actions system and action refinement in the development of parallel systems", In Proc. of CONCUR'91, 298-316, LNCS 527, 1991.

....organization of the architecture. Their work shows that the data parallel programming model can make a clear separation between the programming model and the execution model. Another school of thoughts evolved from the algebraic approach often referred as modularity or compositionality principle [7], 9] Although there is an agreement with the idea that initial design should ignore architectural details, the compositionality principle states that a (parallel or distributed) system can be derived from specifications of its components, by ignoring their internal algebraic structure. However, ....

....Francez [3] Communication Closed Layers principle) For the layer composition F ffl G, any event f from F precedes any event g from G, with which is in conflict, in causal dependence. By comparison, sequential composition, denoted with ; is a representation of the temporal dependence. Work in [7, 9] formulates the algebraic laws of the layer composition, providing that there is no conflict between the programs F and K and between H and G respectively: F [ H) ffl (G[ K) F ffl G) H ffl K) This law is not valid if causal order, expressed by the layer composition, is replaced with ....

[Article contains additional citation context not shown here]

Janssen, W., Poel, M, Zwiers, J.: Actions System and Action Refinement in the Development of Parallel Systems. In Proc. of CONCUR'91, Springer-Verlag, LNCS 527, (1991) 298-316

....One of the typical constraints is that the possible causal relation between two abstract actions is preserved among all the actions of the two implementing processes. For instance, if S = a; b and r(a) a 1 ; a 2 , then the only possible implementation is r(S) a 1 ; a 2 ; b. As pointed out in [14], this can be a serious drawback, because in general a causal relation at an abstract level could be partially forgotten at the concrete one: if only a 1 is to be considered a cause for b, then a 1 ; a 2 jjj b) implements a; b (via r) Some investigations of less rigid forms of action refinement ....

W. Janssen, M. Poel, and J. Zwiers. Actions systems and action refinement in the development of parallel systems. In Baeten and Groote, eds., Concur '91, vol. 527 of LNCS, pp. 298--316. Springer, 1991.

.... Design of Distributed Systems Wil Janssen, University of Twente The design of parallel and distributed systems should start with an initial design stage without premature commitment to machine architecture, followed by a transformational development towards an implementation on a given architecture[CM88, BS91, Ser90, JPZ91, JZ92b, JZ92a, PZ92]. The idea that initial designs should disregard architectural considerations is something we agree with, yet does not lead us to reject the compositional style of reasoning. Rather than blaming the compositional or algebraic approach, we claim that bias towards certain architectures can often be ....

....SdR89, Sto89, Loy92] One of the main advantages of this setup is that transformation techniques for parallel programs like the principle of Communication Closed Layers (CCL) as introduced by Elrad and Francez [EF82] can be put in the form of an algebraic law. This law, which we introduced in [JPZ91], states that: P k R) ffl (Q k S) P ffl Q) k (R ffl S) provided that S is independent from P and R is independent from Q. A similar looking rule where layered composition has been replaced by sequential composition would be not valid, although the two sides are seen to be equivalent in ....

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. In Proceedings of CONCUR '91, LNCS 527, pages 298--316. Springer-Verlag, 1991.

....other on an abstract level, their implementations (as described by the refinement) are also strictly sequential. However, this has turned out to be too restrictive in practice: it might as well be the case that some parts of the implementation of sequential actions overlap in time. As proposed by [14], one way to overcome this problem is to combine action refinement with a dependency relation on actions (in the sense of Mazurkiewicz [16] the ordering among abstract actions is only inherited to dependent parts of their refinements. Dependency based action refinement allows an overlapping ....

.... denotational model [11, 3, 15, 8, 25, 3] The research reported in this paper was partially supported by the Human Capital and Mobility Cooperation Network EXPRESS (Expressiveness of Languages for Concurrency) For dependency based refinement, only the latter approach has been followed yet [14]. In general, the two approaches do not coincide (for a comparison see [12] Operational semantics for action refinement, being consistent with a denotational model, have been defined in [6, 9, 22] however all of them generate transition systems with an enhanced labelling of transition (not just ....

W. Janssen, M. Poel, and J. Zwiers. Actions systems and action refinement in the development of parallel systems. In Baeten and Groote, eds., Concur '91, vol. 527 of LNCS, pp. 298--316. Springer, 1991.

....the other then a compiler is free to reorder them. There is a similar connection to serialisability in data bases. The idea of a dependency relation over the actions of a system can already be found in trace theory as developed by Mazurkiewicz [14] Zwiers et al. have also exploited this idea in [13, 21, 10]. In both cases however, the models used are lineartime, which is to say that the points in time at which choices are made are not represented in the model. We aim at extending this idea to branching time semantics, in particular to strong bisimulation. In this effort we are guided by an existing ....

.... rindA rreq B rindA rreq B rcnf B dreq A rindA rreq B dindB rcnf B 5. 2 Communication Closed Layers An algebraic law that has been applied quite successfully in a linear time setting is the communication closed layers law (CCL) advocated for instance by Zwiers et al. in [13, 21, 10]. As mentioned in the introduction, this law has actually been one of the motivations for the present work. In our setting, CCL can be formulated as follows: 0 (B 1 jj A1 C 1 ) Delta (B 2 jj A2 C 2 ) 1 A = 0 B 1 Delta B 2 1 A jj A1[A2 0 C 1 Delta C 2 1 A : 1) Each ....

[Article contains additional citation context not shown here]

W. Janssen, M. Poel, and J. Zwiers. Actions systems and action refinement in the development of parallel systems. In J. C. M. Baeten and J. F. Groote, editors, Concur '91, volume 527 of Lecture Notes in Computer Science, pages 298--316. Springer-Verlag, 1991.

....which are parallel may be seen to be in fact not parallel. Notice again a problem with names, since the word conflict which we use in relation to resources, is often used for the idea of a choice, in a specification, between different possible behaviours. Ideas similar to ours occur in [God93] [Zwi91]. With our definition, the refined system may have a much richer structure than the unrefined one. Thus we can study such issues as efficiency in time and resources. This implies that at each stage of refinement it is necessary to prove again, if possible, that desired properties of the system are ....

....X and on U a is defined as OE a . It is then clear that the inclusion X Y induces a refinement. 6 Comparisons As we remarked in the introduction, our notion of refinement differs markedly from notions currently being considered in Petri nets and process algebra; rather, it is in the spirit of [CM88, AL87, Zwi91]. The definition which is conceptually closest to our approach is the broader definition of Petri net morphism given in [MM90] where a single Petri net transition can be mapped to an entire computation, possibly composed by many parallel steps. However, due to the freedom with respect to the ....

W. Janssen, M. Poel, J. Zwiers. Action systems and action refinement in the development of parallel systems. In J.C.M. Baeten, J.F. Groote, editors, CONCUR 91, number 527 in LNCS, 1991.

No context found.

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. Technical Report 93-14, University of Twente, 1993.

No context found.

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. In Proc. of CONCUR '91, pages 298--316. Springer-Verlag, LNCS 527, 1991.

....development. After having obtained a sufficient level of detail the resulting design is then parallelized to meet the desired architecture, and finally implemented. Examples of this approach are the work by Elrad and Francez [6] by Chou and Gafni [5] Stomp [19] and by Janssen, Poel and Zwiers [11, 10]. A second strategy would be the parallel strategy, where at the initial stages one is not concerned about architectural issues but simply views the system as a set of continuously executing actions. In the later stages one then introduces architectural decisions as optimizations, leading to an ....

....observation to formulate the so called communication closed layers principle that states that under certain circumstances a layered and a parallel program are equivalent. This principle has later been used in a transformational setting by, amongst others, Stomp [19] and by Janssen, Poel and Zwiers [11, 10]. Based on these ideas we can formulate the following layered strategy. L1) Given a specification or initial design (that is correct by definition) repeatly decompose the specification in sequence of (abstract) actions, using standard techniques as known for sequential program development. ....

[Article contains additional citation context not shown here]

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. In Proceedings of CONCUR '91, LNCS 527, pages 298--316. Springer-Verlag, 1991.

No context found.

W. Janssen, M. Poel and J. Zwiers. Action Systems and Action Refinement in the Development of Parallel Systems. Proceedings of CONCUR 1991, LNCS 527, 298--316, 1991.

No context found.

W. Janssen, M. Poel, and J. Zwiers. Action systems and action refinement in the development of parallel systems. In [BG91], pages 298--316, 1991.

No context found.

W. Janssen, M. Poel & J. Zwiers (1991): Action systems and action refinement in the development of parallel systems. In J. C. M. Baeten & J. F. Groote, editors: Proceedings CONCUR 91, Amsterdam, LNCS 527, Springer, pp. 298--316.

No context found.

Janssen,J., Poel,M. and Zwiers,J.: Action systems and action refinement in development of parallel systems. In: CONCUR 91, LNCS 527, pp 298-316, 1991.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC