E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science, pages 1 -- 50. Springer-Verlag, 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of demand. Pingali and Arvind describe a mechanism for simulating demand driven evaluation in a data flow model; they use data tokens to represent demand [PA85, PA86] Ashcroft describes a system that combines demand flow (via entities called questons) with data flow (via entities called datons) Ash86] 5.2 Snapshots A snapshot is a graph consisting of interconnected labelled nodes. Each node can be viewed as a computational device that responds to a demand for a value by computing that value. Every node has a set of labelled input ports that specify the arguments to the node and a set of ....

E. A. Ashcroft. Dataflow and eduction: Datadriven and demand-driven distributed computing. In J. W. deBakker, W.-P. de Roever, and G. Rozenberg, editors, Current Trends in Concurrency: Overviews and Tutorials, pages 1--50. SpringerVerlag, 1986. Lecture Notes in Computer Science, Number 224.

....a form that is easy to (re)partition. Dataflow is one model of computation that both fits very well with the user model of a monitoring application as a continuously operating network of filters [5] and allows easy partitioning and distribution of a running computation. The operator nets formalism [1], in which an expression of such an application is functionally equivalent to a program in a dialect of the intensional language Lucid [2] fits these requirements well. Here, operators act on sequences of values, in one or more dimensions. A simple operator nets program that computes a running ....

....of the incoming edges for a vertex is significant, the edges are numbered. If results of partial computations are labelled , and , then the following equations specify the mathematical semantics of the program (for a detailed description of semantics of intensional operators refer to [1]) In a monitoring application, the input variables may be the values of two Simple Network Management Protocol (SNMP) Management Information Base (MIB) 4] variables with identical object identifiers on different ....

[Article contains additional citation context not shown here]

E. Ashcroft. Dataflow and Eduction: Data-driven and Demand -driven Distributed Computation. In Current Trends in Concurrency, LNCS 224, pages 1--50. Springer-Verlag, 1986.

....evaluation, both of which avoid unbounded superfluous evaluation. We consider the effects of these extensions on a shared memory multiprocessor implementation of GLU using idealized programs. 1 Introduction Eduction is a lazy intensional computing model for evaluating Lucid programs [1, 2]. It is lazy because output values are only produced when they are needed and it is intensional because the individual output values can be produced over time and not necessarily in order. Eduction has been the basis of all known implementations of various incarnations of Lucid (such as pLucid, ....

E.A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. Technical Report SRI-CSL-151, Computer Science Laboratory, SRI International, Menlo Park, California 94025, U.S.A, 1986.

....processes and we consider sets of such structures. Structures representing histories specify what data have occurred up to a certain stage of development of the respective processes and how these data are related. We assume that data occur in portions called datons (the word taken from Ashcroft, Ashc 85] each daton with a certain meaning. All datons considered in what follows are assumed to have meanings belonging to a nonempty set Meanings that contains an element standing for unspecified . All sets of datons and the set Meanings are assumed to be members of a universal set Universe (cf. ....

....For example, the operations q 7 fix p (p k q) and q 7 fix p (q k p) are equivalent. 8 Closing remarks The concepts we have introduced form a framework in which reactive systems can be defined and analysed. Classical data flow machines are examples of reactive systems (cf. Kahn 74] BA 81] Ashc 85] and [SN 85] A particular feature of such machines is that they may be combined into data flow machines by connecting them with the aid of simple communication channels capable of storing information for later processing. The reactive systems considered in the present paper can also be ....

Ashcroft, E.A., Data Flow and Eduction: Data-Driven and Demand-Driven distributed computation, ESPRIT/LPL Advanced School on Current Trends in Concurrency, Noorwijkerhout, June 10-21, 1985, The Netherlands

....of demand. Pingali and Arvind describe a mechanism for simulating demand driven evaluation in a data flow model; they use data tokens to represent demand [PA85, PA86] Ashcroft describes a system that combines demand flow (via entities called questons) with data flow (via entities called datons) Ash86] 5.2 Snapshots A snapshot is a graph consisting of interconnected labelled nodes. Each node can be viewed as a computational device that responds to a demand for a value by computing that value. Every node has a set of labelled input ports that specify the arguments to the node and a set of ....

E. A. Ashcroft. Dataflow and eduction: Datadriven and demand-driven distributed computing. In J. W. deBakker, W.-P. de Roever, and G. Rozenberg, editors, Current Trends in Concurrency: Overviews and Tutorials, pages 1--50. SpringerVerlag, 1986. Lecture Notes in Computer Science, Number 224.

....input traces to output traces. Differential difference equations and Mealy Moore Machines [11, 12] are typical on line models. We have advocated instead a formal approach to modeling a robotic system consisting of the plant, the controller and the environment. Influenced by the Operator Net model [3] and the Temporal Automaton Model [7] we have developed an algebraic model, called Constraint Nets (CN) 20] of general dynamic systems. The motivation for this unified modeling framework is to provide concise and formal semantics for integrated systems with multiple components and mixed types ....

E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science. Springer-Verlag, 1986.

....can be connected by wires to form a new temporal automaton. A temporal automaton with empty input entities defines a closed system, it otherwise defines a causal system. CHAPTER 7. SUMMARY AND RELATED WORK 94 Operator Nets The Operator Net model is a generalized deterministic dataflow model [Ash86]. A graphical language is defined that is syntactically extremely simple and that is mainly uninterpreted, i.e. using operator symbols rather than particular operators. This uninterpreted graphical language can then be interpreted in several different ways, by starting with different (continuous) ....

E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science, pages 1 -- 50. Springer-Verlag, 1986.

....while subsuming the per pixel processing in sequential functions. The GLU program composition for the raytracing application is struct INPUT int xdimension; int ydimension; int grainsize; char name[64] typedef struct INPUT INPUT; struct STRIP int cols, rows; unsigned char line[cols] rows][3]; typedef struct STRIP STRIP; local int display( string, int, int, int, int, int, int, STRIP ) STRIP raytrace( int, int, int, int, int, int, int, int, int ) local INPUT input( int ) local int xdimin( INPUT ) local int ydimin( INPUT ) local int grainsizein( INPUT ) local string ....

....3 Parallel Program Generation We have shown how parallel applications can be described in GLU at a high level of abstraction where both expression and exploitation of parallelism are implicit. Abstractly, a GLU application can be evaluated using a parallel model of computation called eduction [3, 4]. For each high level application description, the GLU system generates a parallel program that when executed on the target system will eductively discover and harness programinherent parallelism. The generated parallel program, which reflects a particular embodiment of eduction, is referred to as ....

E.A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In Current Trends in Concurrency. Springer Verlag, 1986. Lecture Notes in Computer Science (224).

....computers are suited for exploiting a relatively coarse granularity of parallelism, we will assume that granularity of parallelism in GLU programs is correspondingly coarse. Latent parallelism in GLU programs can be discovered at runtime by an intensional model of computation called eduction[4] from the Lucid part of the program. Eduction corresponds to tagged demanddriven evaluation one that avoids wasteful computation at the expense of evaluation speed. However, when evaluating GLU programs with coarse granularity of parallelism, it is far more important to avoid wasteful ....

E.A. Ashcroft. Dataflow and Eduction: Datadriven and demand-driven distributed computation. In Current Trends in Concurrency. Springer Verlag, 1986. Lecture Notes in Computer Science (224).

....While a GLU program does not prescribe a model of computation, its mathematical semantics has primacy over all possible models of computation that can be used to derive its meaning operationally. The model of computation embodied by the generator executor abstract architecture is called eduction [1]. Eduction corresponds to a lazy evaluation strategy for Lucid (and GLU) programs. The strategy is typically implemented using demand driven execution although it is also possible to use a combination of demand driven and data driven execution [6] Using a lazy model of computation such as ....

E.A. Ashcroft. Dataflow and Eduction: Data-driven and demanddriven distributed computation. In Current Trends in Concurrency. Springer Verlag, 1986. Lecture Notes in Computer Science (224).

....considered to be effective for introducing fine grained (and high) parallelism within an object. Dataflow machines have been extensively investigated in the past decade. Hardware prototypes have also been operational (e.g. Shimada87] Several programming languages for dataflow machines (e.g. [Ashcroft86]) and fine grained parallel programming languages (e.g. Ueda85, Steele87] have also been developed. 0 Object A Object B Object C B A Thread B C thread Fig. 1: Multi threaded Objects Interaction. Some of them employ a combination of dataflow concepts and object oriented concepts. Such ....

E. A. Ashcroft. Dataflow and eduction: datadriven and demand-driven distributed computation. In Current Trends in Concurrency, LNCS 224. Springer-Verlag, 1986.

....GLU compositions express parallelism implicitly. This unburdens the programmer from having to explicitly manage parallelism on a per application basis. The parallelism in GLU compositions is exploited using a novel model of execution namely, demand driven execution (also known as eduction) [1, 2]. The demand driven model is general in that all forms of parallelism can be discovered and exploited. Unlike data driven execution, demanddriven execution only causes useful filter functions to be invoked, thereby avoiding superfluous computations. 2.2 Translation of GLU Programs The ....

E.A. Ashcroft. Dataflow and Eduction: Data-driven and demand-driven distributed computation. In Current Trends in Concurrency. Springer Verlag, 1986. Lecture Notes in Computer Science (224).

.... two concerns have influenced the design of the Lucid dataflow language [AFJ 91] Du and Wadge have made use of this correspondence by designing a spreadsheet around an extension of Lucid called Plane Lucid [DW 90a] DW 90b] This spreadsheet uses the principle of eduction to order calculations [Ash 86] Section 3 discusses our perspective on this approach; section 4 outlines the alternative method of eager evaluation. We believe a hybrid approach is necessary; section 5 is devoted to this. Another issue in spreadsheet languages is the question of syntax. Modern spreadsheets usually use one of ....

....semantic constructs found in Lucid can be translated into the spreadsheet idiom using this notation. 1. All trademarks mentioned herein are the property of their respective owners. 2 2. Eduction Eduction is a term used to describe a tagged demand driven model of dataflow computation [Ash 86] Data items are not submitted for calculation until a demand is made for them by the user. Calculating these items may in turn cause other data items to be needed, so they in turn are demanded. No item is actually calculated until all of the data items upon which it is dependent are available. ....

E.A. Ashcroft. Dataflow and eduction: data-driven and demand-driven distributed computation. In Current Trends in Concurrency, G. Goos and J. Hartmanis, eds., pp 1-50. LCNS 224. Springer-Verlag. 1986.

.... capability of a functional language program, due to its referential transparency property, can be efficiently harnessed if an intensional computing model is used [14, 15] Moreover, this efficient fault tolerance can be naturally and effectively realized using a demand driven implementation [16]. The one main property of functional languages that seems to give opportunity for both spatial and temporal redundancy is referential transparency. Referential transparency ensures that, in general, two occurrences of the same subexpression in a program denote, or will evaluate to, the same ....

....Computing Models Computing models for evaluating functional programs can be divided into two categories: extensional and intensional. Examples of extensional computing models are parallel graph reduction [19, 20] and term rewriting [21] Examples of intensional computing models are eduction [16, 22] and dataflow [23, 24, 25] Extensional Computing Models The most commonly used extensional computing model is reduction. While we restrict the following discussion to reduction, it is equally valid for other extensional computing models. Reduction considers the nonpointwise operations as ....

[Article contains additional citation context not shown here]

E.A. Ashcroft. Dataflow and eduction: Datadriven and demand-driven distributed computation. Technical Report SRI CSL-151, Computer Science Laboratory, SRI International, Menlo Park, California 94025, 1986.

No context found.

E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science, pages 1 -- 50. Springer-Verlag, 1986.

No context found.

E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science, pages 1 --

No context found.

E. A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In J. W. deBakker, W.P. deRoever, and G. Rozenberg, editors, Current Trends in Concurrency, number 224 in Lecture Notes on Computer Science, pages 1 -- 50. Springer-Verlag, 1986.

No context found.

E.A. Ashcroft. Dataflow and eduction: Data-driven and demand-driven distributed computation. In Current Trends in Concurrency. Springer Verlag, 1986. Lecture Notes in Computer Science (224).

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