James R. Russell. Full abstraction for nondeterministic dataflow networks. Annual Symp. Foundations of Computer Science, pages 170--177, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....for any language components c 1 and c 2 and for any context C , the following holds. D(c 1 ) D(c 2 ) IO(C [c 1 ] IO(C [c 2 ] This approach has been used for many languages. For example, it has been used to define fully abstract models of nondeterministic dataflow networks in [Jon89] and [Rus89]. It is not, however, applicable for object oriented languages as it is not always easy to describe a class in terms of a relation between input and output. Classes do not necessarily have any inputs or outputs, in the traditional sense, nor do they necessarily have a final state which could be ....

Russell, J.: Full abstraction for nondeterministic dataflow networks. Technical Report TR 89-1022, Department of Computer Science, Cornell University, Ithaca, NY, 1989.

....model is based on trace sets, which are sets of possible interactions between a process and its environment. Kok s model turned out to be equivalent. They showed that this model is fully abstract for indeterminate dataflow networks with a fair merge primitive, which was then shown by Russell [35] to hold even for dataflow networks with the weakest nondeterministic primitive, bounded choice . Rabinovich and Traktenbrot analyzed the same issues from the point of view of finite observations and came up with general conditions under which a Kahn like principle would hold [32, 33, 34] 3 A ....

RUSSELL, J. R. Full abstraction for nondeterministic dataflow networks. In FOCS'89, pp. 170--176.

....of output values and when a more detailed mathematical model is required. Brock and Ackerman [17] proved that if the nondeterministic primitive fair merge was added to the calculus it was no longer possibly to give a compositional semantics in the history model. As we recall in Sec. 8.1. 1 Russell [116] later showed that the history model is insu#cient also if just the simpler [102] primitive non deterministic choice is added to the calculus. A solution to the problem of giving a compositional semantics for non deterministic dataflow was provided by Jonsson [68] and independently Kok [78] who ....

....semantics for dataflow with fair merge. The model of Jonsson represents the behaviour of networks by sets of (possibly infinite) sequence of input output events, each sequence recording in which order values are read and produced on the ports during a particular, completed computation. In [116] Russell shows that the model of Jonsson is fully abstract even for the calculus with just non deterministic choice and no fair primitives. Although giving a fully abstract semantics, the solution based on language models has left the conceptually simple view of networks in Kahn s semantics in ....

[Article contains additional citation context not shown here]

James R. Russell. Full abstraction for nondeterministic dataflow networks. In FOCS'89, pages 170--176, 1989.

....synchronous dataflow networks. In [8] Barendregt et al. present a model of computable processes which is essentially a model of BNA; but a slightly different choice of primitive operations and constants is used. It is also worth mentioning that the examples of Brock and Ackermann [9] and Russell [10] demonstrating a time anomaly in asynchronous dataflow networks are presented in a concise way in this paper, using network algebra for describing the networks of cells and wires and using process algebra for describing the atomic cells. The paper starts with an outline of network algebra ....

....flows of data which go in one direction at branchings. 3 Asynchronous dataflow reflects this idea better. In order to distinguish between the branching constants with these different interpretations, different symbols are used. Dataflow networks have been extensively studied, see e.g. [3, 4, 5, 8, 9, 10, 12, 13]. 3 Process algebra preliminaries This section gives a brief summary of the ingredients of process algebra which make up the basis for the process algebra models presented in Sections 4 and 5. We will suppose that the reader is familiar with them. Appropriate references to the literature are ....

[Article contains additional citation context not shown here]

J. Russell. Full abstraction for nondeterministic dataflow networks. In FoCS '89. IEEE Computer Science Press, 1989.

.... compositional semantics in a model without causality information to a class of network with the nondeterministic primitive fair merge [BA81] and later Russell has shown with a very simple example that this is true also for networks with the weakest nondeterministic primitive, bounded choice [Rus89]. This is the content of Section 4. Since the details of this later work is still in progress, we mainly describe the background on dataflow networks, only giving the definition of the profunctor model we suggest, describing how its categorical structure is closely related to an existing model of ....

....on different ports. In other words, it provides an example of a class of systems for which independence models are crucial. Jonsson [Jon89] has given a model based on trace sets, which is proven to be fully abstract for indeterminate dataflow networks with a fair merge primitive, and by Russell [Rus89] even for dataflow networks with the weakest nondeterministic primitive, bounded choice 10 . We include a model similar to the model of Jonsson in the end of next section. First we will take a look at profunctors. 4.3 Dataflow as profunctors The bicategory of profunctors is described (using ....

James R. Russell. Full abstraction for nondeterministic dataflow networks. In FoCS, pages 170--176, 1989.

....semantics and the resumptions model of Hennessy and Plotkin [5] The idea of using sequences or traces of some kind to model the behavior of concurrent programs is widespread. For instance, several authors have used traces to build models of determinate or indeterminate dataflow networks, notably [7, 8, 14]. Indeed, others have also used sequences of pairs of states [3, 6, 12] in imperative settings. However, in these papers a pair of states represents a single atomic action while in our model it represents a finite sequence of atomic actions. The semantics presented in [3, 6] are for different ....

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30 th Annual Symposium on Foundations of Computer Science, pages 170--177. IEEE Press, 1989.

....dataflow reflects the idea of intermittent flows of data which go in one direction at branchings well. This idea corresponds to the split merge interpretation. We will use the symbols ffl and ffl for split and merge, respectively. Dataflow networks have been extensively studied, see e.g. [15, 16, 18, 19, 20, 22]. Report No. 98, October 1997 UNU IIST, P.O. Box 3058, Macau Process algebra preliminaries 6 4 Process algebra preliminaries This section gives a brief summary of the ingredients of process algebra which make up the basis for the process algebra models presented in the following sections. We ....

J. Russell. Full abstraction for nondeterministic dataflow networks. In FoCS '89. IEEE Computer Science Press, 1989.

....results in agreement with token pushing semantics. 4 Resolution of the Anomalies In this section, we briefly examine the way in which Keller Brock Ackerman type anomalies are avoided by our semantics for networks. We consider a particularly simple example of such an anomaly, due to J. Russell [31]. Suppose P is a process, having one input channel and one output channel, that obeys the following intuitive algorithm: Nondeterministically choose either: 1) read an input value, then output 0 followed by 1, or (2) output 0, read an input value, then output 0. Let Q be a similar process that ....

J. Russell. Full abstraction for nondeterministic dataflow networks. June 1989. Unpublished manuscript, Cornell University.

....happen with a recursive program, then we have to take the lub of the stores that are produced by the finite unwindings of A on a. 4.2 The problems We turn now to Probabilistic cc. The input output relation is not compositional. The following example is inspired by Russell s simplified version [48] of the BrockAckermann anomaly [15] Example 4.1 Consider the programs A 1 = S 1 ; if R 1 then S 2 ; if R 1 R 2 then S 3 A 2 = S 1 ; if R 1 then S 2 ; if R 1 R 2 then S 4 A 3 = if R 1 then S 1 ; if R 1 R 2 then S 2 S 4 A 4 = if R 1 then S 1 S 2 ; if R 1 R 2 then S 4 A 5 = if R 1 then S ....

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30th Annual Symposium of Foundations of Computer Science, pages 170--177, 1989.

....happen with a recursive program, then we have to take the lub of the stores that are produced by the finite unwindings of A on a. 4.2 The problems We turn now to Probabilistic cc. The input output relation is not compositional. The following example is inspired by Russell s simplified version [52] of the Brock Ackermann anomaly [16] Example 4.1 Consider the programs A1 = S1 ; if R1 then S2 ; if R1 R2 then S3 A2 = S1 ; if R1 then S2 ; if R1 R2 then S4 A3 = if R1 then S1 ; if R1 R2 then S2 S4 A4 = if R1 then S1 S2 ; if R1 R2 then S4 A5 = if R1 then S1 ; if R1 R2 then S2 S3 S i ; R ....

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30th Annual Symposium of Foundations of Computer Science, pages 170--177, 1989.

....with arbitrary determinate processes and as many copies of M as is needed with the same IO relation as N . There is a crucial point to be made here. The notion of IO Relation is not compositional. This is of course the well known Brock Ackerman anomaly [6] The following example, due to Russell [38], is more perspicuous than the original Brock Ackerman example and relies on weaker primitives as we shall see later. Consider a process, shown in the left hand side of figure 3, with a single input channel and a single output channel; ignore, for the moment, the dotted line. Its behaviour is as ....

....propositions are due to Lynch and Stark [21] Proposition 4.12 If M is a determinate automaton, then IO(M) is continuous function. Proposition 4.13 [27] If function f : V 1 ) m (V 1 ) n is continuous, then there exists a determinate automaton M such that IO(M) f . It was shown in [18, 15, 35, 38] that two automata have the same input output relation in every network context iff the closure of their trace sets are the same. This result is known as the full abstraction theorem for trace semantics. 5 The Role of Monotonicity Properties When one is working with determinate automata the IO ....

[Article contains additional citation context not shown here]

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30th Annual Symposium of Foundations of Computer Science, pages 170--177, 1989.

....in operational terms using port automata [14] and show that the Kahn principle is correct with respect to this model. We have generalizations to indeterminate networks [10, 1] and we have expressiveness results that classify the power of different primitives [19, 20, 17] The result of Jonsson [7, 24] shows that traces form a fully abstract semantics for such networks. The most important expressiveness result shows that there is a class of automata, called monotone automata that enjoy nice properties but which exclude simple primitives such as fair merge [19] One of the main results of the ....

....following propositions are due to Lynch and Stark. Proposition 6 [14] If M is a determinate automaton, then IO(M) is continuous function. Proposition 7 [18] If function f : V 1 ) m (V 1 ) n is continuous, then there exists a determinate automaton M such that IO(M) f . It was shown in [12, 7, 23, 24] that two automata have the same input output relation in every network context iff the closure of their trace sets are the same. This result is known as the full abstraction theorem for trace semantics. 3 Monotone Automata and Monotone Trace Languages In this section we state and prove the ....

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30th Annual Symposium of Foundations of Computer Science, pages 170--177, 1989.

....very pleasant application of Scott s semantic ideas. In almost any elaboration of Kahn s model, the situation becomes much more difficult. In the context of indeterminate dataflow, recent work by Kok, Jonsson and others has shown that one gets fully abstract models from the traces of computations [9, 11, 17]. Traces do not, however, give one the same level of abstraction that is provided by being able to think of processes as functions. Similarly, though process algebra has now reached a high degree of mathematical maturity and elegance, see, for example, the recent books by Milner and by Hennessy ....

J. R. Russell. Full abstraction for nondeterministic dataflow networks. In Proceedings of the 30th Annual Symposium of Foundations of Computer Science, Lecture Notes in Computer Science 442, pages 170--177, 1989.

....for any language components c 1 and c 2 and for any context C , the following holds. D(c 1 ) D(c 2 ) IO(C [c 1 ] IO(C [c 2 ] This approach has been used for many languages. For example, it has been used to define fully abstract models of nondeterministic dataflow networks in [Jon89] and [Rus89]. It is not, however, applicable for object oriented languages as it is not always easy to describe a class in terms of a relation between input and output. Classes do not necessarily have any inputs or outputs, in the traditional sense, nor do they necessarily have a final state which could be ....

Russell, J.: Full abstraction for nondeterministic dataflow networks. Technical Report TR 89-1022, Department of Computer Science, Cornell University, Ithaca, NY, 1989.

....dataflow networks. In [10] Barendregt et al. present a model of computable processes which is essentially a model of BNA; but a slightly different choice of primitive operations and constants is used. It is also worth mentioning that the examples of Brock and Ackermann [18] and Russell [40] demonstrating a time anomaly in asynchronous dataflow networks are presented in a concise way in this paper, using network algebra for describing the networks of cells and wires and using process algebra for describing the atomic cells. The paper starts with an outline of network algebra (Section ....

....this model is left open. The relationship between the branching constants from both kinds remains to be investigated as well: the copy equality test interpretation of the branching connections seems also meaningful in the asynchronous case. Dataflow networks have been extensively studied, see e.g. [10, 17, 18, 19, 33, 34, 37, 40]. 3 Process algebra preliminaries This section gives a brief summary of the ingredients of process algebra which make up the basis for the process algebra models presented in Sections 4, 5 and 6. We will suppose that the reader is familiar with them. Appropriate references to the literature are ....

[Article contains additional citation context not shown here]

J. Russell. Full abstraction for nondeterministic dataflow networks. In FoCS '89. IEEE Computer Science Press, 1989.

No context found.

James R. Russell. Full abstraction for nondeterministic dataflow networks. Annual Symp. Foundations of Computer Science, pages 170--177, 1989.

No context found.

J. Russell. Full abstraction for nondeterministic dataflow networks. Technical Report TR 89-1022, Department of Computer Science, Cornell University, Ithaca, NY, 1989.

No context found.

J. Russell. Full abstraction for nondeterministic dataflow networks. In: Proc. FoCS'89, IEEE Computer Science Press, 1989.

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