P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins. Data Driven and Demand Driven Computer Architectures. ACM Computing Surveys, 14(1):93--143, 1982.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and memory locations reserved for operand datatokens. In dynamic data flowarchitectures, instruction cells have nostorage reserved for operands. Instead, each data token is tagged with information that associates it with an instruction cell and with other data tokens executing in the same context [13]. We propose a newdata flowprogram representation scheme which is based on instruction sequences for a prioritized queue machine. An instruction cell shall occupyarange of addresses in twoaddress spaces instruction space and data token space. The operation code and destination offset parts of ....

....to select the set, and then comparing the least significant n bits of the data token address with the key stored in the set. If a match is found, the context becomes enabled for execution. The execution model supported by this machine is likethe communicating sequential process (CSP) model [13] in that contexts are explicitly enabled for execution by the FORK 6 Copyright (c) 1985 by The Institute of Electrical and Electronics Engineers, Inc. instruction. However, once a context is activated, its execution is controlled by the availability of operands and processors (data flow) The ....

Philip C. Treleaven, David R. Brownbridge and Richard P.Hopkins, "Data-Drivenand Demand-DrivenComputer Architecture," Computing Surveys, Vo l. 14, No. 1, March 1982, pp. 93-143. -11-

....and efficiency. A dataflow [27] visual programming model is combined with concepts from object oriented programming to achieve modularity by insulating data representations from dataflow modules. This combination also provides a mechanism for implementing demand driven dataflow (lazy evaluation) [120] in a straightforward manner. Threads [111] are used to provide task parallelism expressed naturally in the dataflow graph, as well as data parallelism coded explicitly in different components. 1 SCIRun is pronounced ski run and derives its name from the Scientific Computing and Imaging (SCI) ....

....A dataflow [27] visual programming model is combined with concepts from object oriented programming to achieve higher modularity by insulating data representations from dataflow modules. This combination also provides a mechanism for implementing demand driven dataflow (lazy evaluation) [120] in a straightforward manner. Threads [111] are used to provide task parallelism expressed naturally in the dataflow graph, as well as data parallelism coded explicitly in different components. Finally, we extend the concept of dataflow s internal communication by allowing more flexible ....

TRELEAVEN, P., BROWNBRIDGE, D., AND HOPKINS, R. Data-driven and demand-driven computer architecture. Computing Surveys (March 1982), 93--143.

....complete axiomatisation of the bisimulation equivalence is also given. Keywords: Multiprocessors, Distributed systems, CCS 1 Introduction A number of different types of parallel and distributed machines have been built. These include vector machines [Rus78] data driven demand driven machines [TBH82], shared memory systems, like the Sequent, and distributed memory systems ranging from a network of work stations to well structured architectures like the hypercube. More parameters such as heterogeneity, replication for fault tolerance etc. add to the variation. Despite the variety of machines, ....

P. Treleaven, D. Brownbridge, and R. Hopkins. Data Driven and Demand Driven Computer Architectures. ACM Computing Surveys, 14(1), 1982.

....are closely connected via very high speed communication links. Physically distributed computers connected by slow message passing networks (e.g. ARPANET) are usually excluded. CHAPTER 2. THE QUEST FOR HIGH PERFORMANCE 37 been developed. Dataflow architectures [Gurd85] and graph reduction machines [Trel82] are examples of such paradigms. 2.5.3 Parallel Programming Models and Languages The von Neumann model of computation is typically characterized as control driven. There is a single thread of control normally passed sequentially from instruction to instruction thus determining the sequence in ....

Treleaven, P.C., Brownbridge, D.R., Hopkins, R.P., "Data-Driven and Demand-Driven Computer Architecture", ACM Computing Surveys, 14, 1, January 1982, pp. 93-143.

....Many novel architectures for exploiting parallelism have been proposed and built over the past twenty years, ranging from array and vector processors through to more recent dataflow, graph reduction and connection machines. Those interested in further details will find a useful survey in [3] and some more recent introductory articles in [4] # PARALLEL PROCESSING IN DOCUMENT FORMATTING 27 PARALLELISM RECOGNITION AND IMPLEMENTATION For any problem to be readily amenable to parallel processing there are certain requirements which bear on the nature of the problem itself, the ....

Philip C. Treleaven, David R. Brownbridge, and Richard P. Hopkins, `Data-Driven and Demand-Driven Computer Architecture', ACM Computing Surveys, 14 (1), (1982).

....among others. ffl The iteration level distinguishes between different iterations in a loop. ffl The index specifies a member within a structure, such as an array. The communication between major components in a dataflow machine is carried through packets. In the packet communication organization [TBH82], a packet, such as a token, represents a unit of work that is consumed by a system component such as processors, which in turn may produce more packets to be consumed by other components. There might be several different types of packets in the system. The program execution can hence be viewed as ....

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins. Data-driven and demanddriven computer architecture. ACM Computing Surveys, 14(1):93--144, 1982.

....paradigms, illustrates them using common examples, and promulgates some simple methodological ideas about using these concepts to iterate to efficient parallel programs. 2.2. 3 Further Classification In addition to the earlier analysis, we repeat here a classification which relates to those of [Treleaven 85] and [Almasi 89] where the study was predominantly focussed on socalled MIMD (multiple instruction stream, multiple data stream) computers. This classification produces a two way table (Figure 2.4 below) as a result of distinguishing the nature of the control mechanism and logical address ....

P Treleaven, Control-Driven, Data-Driven and Demand-Driven Computer Architectures, (abstract), Parallel Computing 2, 1985.

....none is absolutely suitable. Due to its elegance and simplicity, the pure dataAEow model has been the subject of many research eoeorts. Since the early 1970s, a number of dataAEow computer prototypes have been built and evaluated, and dioeerent designs and compiling techniques have been simulated [17,66,88,151,168,185,196,205,208,216,219]. Clearly, an architecture supporting the execution of dataAEow graphs should support the AEow of data. Depending on the way of handling the data, several types of dataAEow architectures emerged in the past [15] single token per arc dataAEow [72] tagged token dataAEow [23,229] and explicit ....

P.C. Treleaven, D.R. Brownbridge, and R.P. Hopkins, Data-driven and demand-driven computer architectures, Computing Surveys, 14 (1982), pp. 93143.

....the level of granularity of the model (and thus its size and complexity) can be dynamically adjusted, according to the needs of the design space exploration. Table 1 summarizes the various types of nodes supported in the algorithmic model and the firing rules defined for these node types. 13][14] Three types of firing rules are defined: input conjunctive output conjunctive, input conjunctive output disjunctive, and input disjunctive output conjunctive. Nodes with input conjunctive output conjunctive firing rules become enabled for execution only when all incoming edges have data ....

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins, "Data driven and demand driven computer architecture," ACM Comput. Surv., Mar. 1982.

....are based upon the number of programs, number of execution states and the number of work spaces. He describes architectures by a three letter code with each letter either S or M depending upon whether single or multiple instances of the characteristic are supported. Treleaven s classification [195] is based on the observation that any architecture is a combination of the control and data mechanisms. The various control mechanisms include: control driven, data driven, demand driven, pattern driven. The data mechanisms include: shared memory, message passing. This classification covers a ....

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins. Data Driven and Demand Driven Computer Architectures. ACM Computing Surveys, 14(1):93--143, 1982.

....with input and output data, and a computational step could be well applied to reactive systems. That is to say that a computational step is invoked only when the required input data is present. This principle is reminiscent of the dataflow architecture with a data driven computing organization [TB82], where the computational step is executed once the necessary data becomes available. The computational step itself may represent the execution of one or more instructions, or in other words, a subroutine of arbitrary size, which is concerned with pure computations. The above observation leads to ....

P. C. Treleaven, D. R. Brownbridge, R. P. Hopkins, "Data-Driven and DemandDriven Computer Architecture", in Dataflow and Reduction Architectures, Editor S. S. Thakkar, Computer Society Press of the IEEE, pp.4-54, 1987.

....afore mentioned. Now we consider the four aspects that drove the ALFA design process: language choice, control philosophy, memory latency, and communication mechanism. 3.1. Language Choice In the past, a clear distinction has been done between the data driven and demand driven execution models [TRE82]. However, in these years this distinction has become less and less evident. In fact, Davis and Keller [DAV82] observed that data driven execution is like demand driven execution in which all data have already been requested, while Wei and Gaudiot [WEI88] presented a demand driven evaluation ....

Treleaven P., Brownbridge D.R., and Hopkins R.P. Data-Driven and Demand-Driven Computer Architectures. ACM Computing Surveys, March 1982.

....from the user program and convert the program into a set of chare definitions. Currently, a parallel Prolog interpreter and a compiler have been developed by using the chare kernel language [14, 15, 25] 2. Related Work The ideas developed within the dataflow approach are relevant to our work [1, 30]. All machine details are hidden in this model. A program written in a high level functional language is translated to dataflow graphs. Each node in the dataflow graph represents an individual instruction and is triggered by the arrival of relevant data. Obviously, it is a fine grain approach. ....

Treleaven, P. C., Brownbridge, D. R., and Hopkins, R. P. Data-driven and demanddriven computer architecture. ACM Computi. Surv., 14(1):93--143, March 1982.

....we are representing query operations by objects and hence adopt the object oriented paradigm to describe query processing. Since objects are processes, we are able to execute queries in parallel and distributed by means of interaction. Our schema of execution resembles graph reduction techniques [TBH82] The computation of terms is postponed until necessary, which is known as lazy evaluation. A similar approach to query processing was presented in [BFN82] The query system is conceptually an object society of its own. It is defined using the same concepts being used for any other object ....

....each node of the operator graph is transformed into a query object, we achieve a structure of query objects to process a query. This corresponds to the representation of an expression by a graph connecting subexpressions by pointers. It is then possible to employ a graph reduction technique [TBH82] for query evaluation. The graph reduction technique introduces horizontal parallelism whenever it is possible. The execution is demand driven, i.e. intermediate results are evaluated only when they are needed. The evaluation of terms is postponed until necessary (lazy evaluation) The technique ....

Treleaven, P. C.; Brownbridge, D. R.; Hopkins, R. P.: Data-Driven and Demand-Driven Computer Architecture. ACM Computing Surveys, Vol. 14, No. 1, 1982, pp. 93--143.

....the pipeline is long. This also delays error detection if an error occurs in the data set. The coarse grain approach reflects the traditional firing rule in a pure dataflow system which mandates that all required inputs be available on input arcs and that the output arc have space for the result [98]. Landry [56] modified the conditions for the firing rule and proposed the Firing Semantic Set (FSS) which leads to more general and flexible firing condition specifications. An FSS is a specification of the set of input and output configurations that allow a node to be enabled. In the SDTM ....

P.C. Treleaven, D.R. Brownbridge, and R.P. Hopkins. Data-driven and demand-driven computer architecture. Computing Surveys, 14(1):93--143, March 1982.

....architectures characteristic of the mid 80 s. Many models broadly comparable to the Flagship Machine model were being explored and implemented at that time. We cite [Fasel and Keller (1986) FPLCA] Woods (1986) as containing representative collections of papers from that period. See also [Treleaven et al. 1982)] The present study differs from these in two important respects. Firstly the present work is a self contained theoretical enterprise (though obviously strongly influenced by considerations of practice) while most of the cited work has a strongly implementation oriented flavour. Secondly, apart ....

Treleaven P.C., Brownbridge D.R., Hopkins R.P., Data Driven and Demand Driven Computer Architecture. Computing Surveys 14, 93-143, (1982).

.... [34] Parallelism annotated by the programmer is used in the h; Gi machine [3] the GAML machine [44] the PAM machine [43] the PABC machine [50] and the GRIP machine [55, 54] A survey of these recent designs may be found in [36] Early parallel graph reduction machines have been described in [59, 32, 64]. 2.1 Conditions for successful job based reduction Any expression can be annotated, but parallel evaluation is only beneficial if the jobs satisfy certain constraints (the so called job conditions) a) A job has to be self contained, that is the runtime representation of the job must not ....

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins. Data-driven and demand-driven computer architecture. ACM computing surveys, 14(1):93--142, Mar 1982.

No context found.

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins. Data Driven and Demand Driven Computer Architectures. ACM Computing Surveys, 14(1):93--143, 1982.

No context found.

P. C. Treleaven, D. R. Brownbridge, and R. P. Hopkins, "Data-Driven and DemandDriven Computer Architecture," ACM Computing Reviews, Vol.14, No.1, March 1982, pp.93--143.

No context found.

P. C. Treleaven, D. R. Brownbridge and R. P. Hopkins, `Data-driven and demand-driven computer architectures', ACM Computing Surveys, 14, (1), 93--143 (1982).

No context found.

P.C. Treleaven, D.R. Brownbridge, R.P Hopkins (1982). Data-driven and demand-driven computer architecture.

No context found.

Treleaven, P. C., D. R. Brownbridge, and R. P. Hopkins, "Data-Driven and DemandDriven Computer Architecture," Computing Surveys, vol. 14, no. 1, pp. 93-143, March, 1982. 245

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