L. Jamieson, "Characterizing parallel algorithms," in The Characteristics of parallel algorithms, L. Jamieson, D. Gannon, R. Douglass (eds.), MIT Press, Cambridge, MA, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....in which the execution times of the tasks follow known probability distributions and communication times are zero or constant. 2 Background 2. 1 Algorithmic Structures It is known that many algorithms possess an identifiable structure and that many algorithms share communication patterns [1]. Figure 1 gives sample precedence graphs for four parallel structures commonly found in algorithms: asynchronous, nearest neighbor, synchronous, and asynchronous master slave. In each of these structures, arcs from one node to another indicate that the first node (task) must complete execution ....

L. H. Jamieson, "Characterizing parallel algorithms," in The Characteristics of Parallel Algorithms (L. H. Jamieson et al., eds.), pp. 65--100, Cambridge, MA: MIT Press, 1987.

....4.2.1 Design Characterization A key requisite for providing useful design feedback is a mechanism for analyzing the design being optimized. Previous research has shown the use of design characteristics, or design properties, for predicting a variety of design metrics of the final implementation [Jam87, Gue94]. Effectiveness of various optimizations can also be linked to a design s properties. The properties identify situations in which certain optimization techniques will work well and can be used in predicting their effectiveness. Classes of properties used in this work are outlined below. Examples ....

Jamieson, L. Characterizing parallel algorithms, in The Characteristics of parallel algorithms, L. Jamieson, D. Gannon, R. Douglass (eds.), MIT Press, Cambridge, Mass., 1987.

....an application task onto a parallel machine. Case studies, such as the one presented here and ones for other types of applications (e.g. 34] are a necessary step in developing techniques for using these characteristics and attributes in software tools that will facilitate the mapping process [12]. Furthermore, understanding and quantifying the interaction of important application compute characteristics [13] with important machine attributes is a key enabling technology needed for facilitating software tools for mixed machine heterogeneous computing, where a task is decomposed into ....

L. H. Jamieson. Characterizing parallel algorithms. In L. H. Jamieson, D. G. Gannon, and R. J. Douglass, eds., The Characteristics of Parallel Algorithms, The MIT Press, Cambridge, Massachusetts, 1987.

....of communication [501, 120, 377, 232, 434] Another consideration is the matching of the communication patterns to the interconnection topology of the hardware. It is desirable to map interdependent tasks onto threads that will execute on PEs that have a direct communication link between them [73, 290, 552]. This consideration was important in first generation message passing machines that used a store and forward mechanism for data transfer, but has lost some of its justification in contemporary systems based on wormhole routing. However, mapping to match the topology can still be important with ....

L. H. Jamieson, "Characterizing parallel algorithms". In The Characteristics of Parallel Algorithms, L. H. Jamieson, D. B. Gannon, and R. J. Douglass (eds.), pp. 65--100, MIT Press, 1987.

....by assigning tasks to different processors. In addition, some of the tasks are subsequently replicated in order to enhance performance even further. In other words, exploiting parallelism is achieved by combining functional (or task oriented) parallelism, with data parallelism by replication [9]. At present, there is hardly support available for developing parallel real time applications, although several methods, techniques, and languages have been developed [6] However, these approaches hardly address the aspect of exploiting parallelism for the sake of meeting performance ....

....models. A structure model is used to express the structure of the application, whereas a behavioral model is used for describing the dynamics of the system. The structure model is a design expressing an application as a collection of activities that act in parallel. This task oriented parallelism [9] may not be sufficient to achieve the required performance. Therefore, ADL also provides several replication techniques effectively leading to incorporating data parallelism in a design. Designing the structure of an application, specifying how data parallelism can be exploited, and modeling the ....

L.H. Jamieson. Characterizing Parallel Algorithms, In L.H. Jamieson, D.B. Gannon and R.J. Douglass, (eds.), The Characteristics of Parallel Algorithms. MIT Press, Cambridge, Mass., 1987.

....how our methodology provides portability of algorithm specification, reusability and ease of expressibility without significantly sacrificing efficiency. 1 Introduction In parallel computing, the set of operations and the partial order in which they may be carried out define an ideal algorithm [10]. This ideal algorithm can be specified without introducing any unnecessary sequentiality in the code by using a form of concurrent objects, i.e. actors. In practice, implementations of a parallel algorithm may not use all the parallelism available in the ideal version of the algorithm, because ....

L. H. Jamieson. Characterizing parallel algorithms. In R. J. Douglass L.H. Jamieson, D.B. Gannon, editor, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....the system, the interconnection network employed, and the technology available for system implementation. In general, many algorithms exist for the solution of particular application problems and, for each such algorithm, a variety of architecture and topology specific formulations may be devised [14]. These architectural and topological embodiments are in turn dependent on the capabilities of the interconnection and packaging technologies available for system implementation. Any attempt to design high performance, cost efficient MPPs must consider the interplay among all these entities an ....

L. H. Jamieson, "Characterizing parallel algorithms, " in The characteristics of parallel algorithms (L. H. Jamieson, D. B. Gannon, and R. J. Douglass, eds.), pp. 65--100, Cambridge: The MIT Press, 1987.

....the processors on the mesh. 2 Our Approach Current programming methods for concurrent computers intermix specification of resource management policies with the code specifying the ideal algorithm i.e. the set of operations in the algorithm and the partial order in which they may be carried out [19]. The conflation of design goals resulting from such intermixing complicates the code and reduces its reusability. This paper discusses an object oriented methodology to support the separate specification of an ideal algorithm and the policies used to partition and distribute it onto a concurrent ....

L. H. Jamieson. Characterizing parallel algorithms. In R. J. Douglass L.H. Jamieson, D.B. Gannon, editor, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....parallel signal processor system consisting of 256 modules. We have embedded classical as well as non standard topologies, such as hypercube, 3 dimensional torus, star graph, a dense graph and randomly generated graphs. 1 INTRODUCTION The need to execute time critical or large scale applications [13] as fast as possible has led to a tremendous growth of interest in parallel architectures offering an enormous computing potential. This move to parallelism has introduced new degrees of freedom to both Copyright c fl1998 by the Association for Computing Machinery, Inc. Permission to make digital ....

....the algorithm designs. One important design criterion is the interconnection topology of multiprocessor systems on one hand and the communication topology of parallel algorithms on the other hand, because a good match between both topologies is essential for an effective parallel processing [20] [13]. There are many works addressing the topological aspect of mapping distributed computations to multiprocessor systems [5] 14] 10] Some of these studies investigate the emulation of a topology T1 on another topology T2 , which is known as the graph embedding problem. It arises in two cases: In ....

L.H. Jamieson, Characterizing Parallel Algorithms, The Characteristics of Parallel Algorithms, MIT Press, 1987, pp. 65-100.

....printers or administrators; new synchronizers may simply be instantiated. 5.2 Modular Specification of Resource Management Policies Expressing a parallel algorithm in terms of primitive actors provides a logical specification of the algorithm. Such a specification may be called an ideal algorithm [8]. The time taken by the ideal algorithm, in the presence of unbounded resources and zero communication cost, is determined by the sequential depth of the longest path in the partial order defined by the actor computation. However, neither of these assumptions is realistic. In particular, ....

....Programming. MIT Press, Cambridge, Massachussets, 1987. AST s generalize and abstract over Concurrent Aggregates ( 7] 7] A. A. Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993. For actor based computer architecture such as J machine, see [8]. 8] W. Dally. A VLSI Architecture for Concurrent Data Structures. Kluwer Academic Press, 1986. 9] S. Matsuoka and A. Yonezawa. Analysis of Inheritance Anomaly in Object Oriented Concurrent Programming Languages. In G. Agha, P. Wegner, and A. Yonezawa, editors, Research Directions in ....

[Article contains additional citation context not shown here]

L. H. Jamieson. Characterizing Parallel Algorithms. In R. J. Douglass L.H. Jamieson, D.B. Gannon, editor, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....for different parallel algorithms. We demonstrate how our methodologyprovides portability of algorithm specification, reusability and ease of expressibility. 1 Introduction In parallel computing a set of operations and the partial order in which they may be carried out define an ideal algorithm [7]. The ideal algorithm may be specified without introducing any unnecessary sequentiality by using maximally concurrent objects, i.e. actors. In practice, limitations on computation and communication resources in practical architectures make implementations of a parallel algorithm use only part of ....

L. H. Jamieson. Characterizing parallel algorithms. In R. J. Douglass L.H. Jamieson, D.B. Gannon, editor, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....enhancement of the performance. Keywords: Configuration, embedding, parallel computing, heterogeneous networks. 1 INTRODUCTION The time to compute large scale applications on massive parallel computers depends crucially on the communication overhead caused by the data exchange between processors [2]. Thus a good match between the data dependencies defined by the application and the topology of the processor interconnection network is essential for an effective parallel processing. An interesting class of interconnection networks are the so called configurable networks [3] 4] which can be ....

L.H. Jamieson, Characterizing Parallel Algorithms, The Characteristics of Parallel Algorithms, MIT Press, 1987, pp. 65-100.

....the implementation but it may influence the solution itself [8] So, the model allows for the hardware description to be constructed and analysed simultaneously with the software. This enables us to model SDMs where the hardware dependent designs are derived from abstract (virtual) designs [13], and also those where the hardware characteristics are taken into account during the design, especially to reduce performance design errors [4] Certainly, an implementation is only derived after a software hardware description has been produced. However, many approaches assume the ....

L. H. Jamieson. Characterizing parallel algorithms. In L.H. Jamieson, D. Gannon, and R. J. Douglas, editors, The characteristics of Parallel Algorithms, pages 65--100. The MIT Press, 1987.

....modifying the code specifying the logic of a parallel algorithm. We illustrate our methodology for parallel algorithms that use dynamic data structures. 1 Introduction In parallel computing, the set of operations and the partial order in which they may be carried out define an ideal algorithm [25]. We use Actors, a form of concurrent objects, to specify ideal algorithms. Actors naturally express an ideal algorithm without introducing unnecessary sequentiality in the code. Data parallelism can be naturally expressed by messages broadcast to a group of actors. Functional parallelism may be ....

L. H. Jamieson. Characterizing parallel algorithms. In R. J. Douglass L.H. Jamieson, D.B. Gannon, editor, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....assuming a certain environment (target machine) So, the main artifact of this activity is the software product itself. The software description activities are concerned with the specification and design of the software product. One approach in the development of HPS is to produce a virtual design [15] which is independent of hardware (portable) but at the sametimepresents the level of performance required. Unfortunately, there is a trade off between portability and performance a specific design which exploits the characteristics of the hardware tends to present the best performance [1] A ....

....implementation but it may also influence the solution itself. For this reason, the model allows for the hardware description to be constructed and analysed simultaneously with the software. This permits modelling SDMs where the hardware dependent designs are derived from abstract (virtual) designs [15], and also where the hardware characteristics are taken into account during the design especially to reduce performance design errors as proposed in [4] The artifacts of the hardware description activities can be specific hardware characteristics or a description of the hardware which can ....

[Article contains additional citation context not shown here]

L. H. Jamieson. Characterizing parallel algorithms. In L.H. Jamieson, D. Gannon, and R. J. Douglas, editors, The characteristics of Parallel Algorithms, pages 65--100. The MIT Press, 1987.

....needed to understand the sources of inefficiency in a given architecture with respect to a given algorithm. An understanding of the interaction between the algorithmic and architectural characteristics of a parallel system can give us such information. Studies undertaken by Kung [16] and Jamieson [13] help identify some of these characteristics from a theoretical perspective but they do not provide any means of quantifying their effects. 1 The term, parallel system, is used to denote an algorithm architecture combination. Parallel algorithms designed for an idealized machine model, project ....

Leah H. Jamieson. Characterizing Parallel Algorithms. In L. H. Jamieson, D. B. Gannon, and R. J. Douglas, editors, The Characteristics of Parallel Algorithms, pages 65--100. MIT Press, 1987.

....and the Office of Naval Research. architectural properties to the greatest extent. This general approach and detailed descriptions of several property measures in the context of ASIC area estimation were presented in [5] Properties have been investigated in related work for performance analysis [6]. In particular, this effort considered properties relevant for analyzing high speed parallel processing designs. The property measures of interest in this paper are of a different nature due to both the type of designs targeted (ASIC) and the performance goal (low power) Total chip power is ....

L. Jamieson, "Characterizing Parallel Algorithms," in The Characteristics of parallel algorithms, L. Jamieson, D. Gannon, R. Douglass (eds.), MIT Press, Cambridge, Mass., 1987.

....programming tools A parallel program has its own characteristic attributes, such as the nature of shared variable access, degree of parallelism, placement of tasks, synchronization between tasks, etc. Techniques for characterization of parallel algorithms have been studied before. e.g. Jamieson [13] used the characteristics of parallel algorithm, in conjunction with the characteristics of parallel architectures, to provide an understanding of how well the algorithm is suited to different architectures. We believe that a knowledge of the characteristics of a parallel program can help make the ....

Leah H. Jamieson. Characterizing parallel algorithms. In Leah H. Jamieson, Dennis Gannon, and Robert J. Douglass, editors, The characteristics of parallel algorithms. The MIT Press, 1987.

....the two is somewhat artificial, except for the fact that processes can be replicated into regular geometrical structures. In this way, ADL also supports data parallelism, although emphasis remains on exploiting parallelism through functional decomposition (also referred to as task parallelism [4, 11]) Behavior of activities is modeled by means of state transition machines (STMs) However, where STMs normally consist of a single notion of a state, and where transitions between states can only occur as the result of an event, ADL uses an approach which more or less combines STMs and flow ....

L.H. Jamieson. Characterizing Parallel Algorithms, In L.H. Jamieson and D.B. Gannon and R.J. Douglass, (ed.), The Characteristics of Parallel Algorithms, chapter 3, pp. 65--100. MIT Press, Cambridge, Mass., 1987.

....or a timeout at that gate. In other words, we need not explicitly model a request for the current fuel level by means of datum or token sent from the requesting activity to the fuel tank. 3.1. 3 Process replication Although a structure and a behavior model reflect inherent, functional parallelism [15] it would be erroneous to state that this parallelism is targeted towards achieving performance. In this sense, the modeling support presented so far captures the traditional support that is provided by most design methods used today. And as we have argued in Section 2.1, we need more when dealing ....

L.H. Jamieson. Characterizing Parallel Algorithms, In L.H. Jamieson and D.B. Gannon and R.J. Douglass, (ed.), The Characteristics of Parallel Algorithms, chapter 3, pp. 65--100. MIT Press, Cambridge, Mass., 1987.

....Many computer vision systems of different sizes and configurations have been proposed and developed. Exploitation of parallelism is one way to achieve computational speedup for a broad range of applications, where the exploitation of data parallelism and or function parallelism is possible [2]. In data parallelism, the same operation is performed over many data elements by many processors simultaneously. Function parallelism allows two or more tasks to be performed simultaneously. Also, the degree of parallelism can be divided into fine grain, medium grain, and coarse grain sizes. ....

L.H. Jamieson, "Characterizing parallel algorithms," in The Characteristics of Parallel Algorithms, MIT press, Cambridge, MA, pp. 65 ¸ 100, 1987

No context found.

L. H. Jamieson, "Characterizing Parallel Algorithms, " in The Characteristics of Parallel Algorithms, L. H. Jamieson, D. Gannon, and R. J. Douglass, editors, M.I.T. Press, Cambridge, MA, 1987, pp. 65-100.

No context found.

L. H. Jamieson, "Characterizing Parallel Algorithms, " in The Characteristics of Parallel Algorithms, L. H. Jamieson, D. Gannon, and R. J. Douglass, editors, M.I.T. Press, Cambridge, MA, 1987, pp. 65-100.

No context found.

L. H. Jamieson, "Characterizing Parallel Algorithms," in The Characteristics of Parallel Algorithms, L. H. Jamieson, D. Gannon, and R. J. Douglass, editors, M.I.T. Press, Cambridge, MA, 1987, pp. 65-100.

No context found.

L. Jamieson, "Characterizing parallel algorithms," in The Characteristics of parallel algorithms, L. Jamieson, D. Gannon, R. Douglass (eds.), MIT Press, Cambridge, MA, 1987.

No context found.

L.H. Jamieson. Characterizing Parallel Algorithms, In L.H. Jamieson and D.B. Gannon and R.J. Douglass, (ed.), The Characteristics of Parallel Algorithms, chapter 3, pp. 65--100. MIT Press, Cambridge, Mass., 1987.

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