D. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, vol. 30, issue 2, pp 123 -- 169, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....makes envisioning the execution of parallel programs very dif cult. One way to increase the usability of the parallel paradigm is to abstract away from the details of explicit parallelism. Various ways of performing this abstraction are reviewed in a recent survey of parallel programming models [ST98] 1.2.1 Implicit Parallelism Implicit parallelism is introduced behind the scenes by a compiler or optimiser so that the programmer is largely unaware of it. There has been surprisingly little work on implicit parallelism despite its usability advantages. The problem of introducing parallelism ....

....(f x) l has a processor farm of the form: x: g (f x) x: g (f x) x: g (f x) map x: g (f x) l Problems with this approach come when the expressiveness of the programming language is diminished because the programmer is forced to use the pre de ned skeletons. Furthermore in [ST98] Skillicorn claims that there is no natural way to develop algorithmic skeleton programs from higher level programming constructs or at the skeleton level. 1.3 Functional Programming In the functional programming paradigm, variable assignments, which are common in imperative programming, are ....

David B. Skillicorn and Domenico Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123-169, June 1998.

....If a parallel code is not portable, then it is just a one time parallelization, and it is interesting only for #t years (see figure 3) 4. FIRST NEXT CHALLENGE: NEED FOR LONG LINES OF PARALLEL COMPUTERS Parallel codes have become portable (see previous section) but as said Skillicorn and Talia [12] execution time [of a parallel program] can change by an order of magnitude when it is moved across architecture families . This is true for sequential computing too, but changes are not so important and it exists some long lines of sequential computers (as Intel PC, Sparc Workstations, ....

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing surveys, 30(2), June 1998.

....between distributed address space and shared address space approaches. These classes are further subdivided based on the communication schemes employed. However, their focus is on programming languages the relatively new field of frameworks and middleware is not cov121 ered. Skillicorn and Talia [166] presented a classification of models and languages for parallel computation which is based upon the level of abstraction of parallelism and communication. In our classification we take a more general approach and distinguish between language based and framework and middleware based systems. ....

....as pthreads [114] or heterogeneous operating systems, such as Legion [92] or Globe [183] are beyond our focus. Similarly, we do not discuss systems that focus on homogeneous architectures, such as Mosix [13] Our selection is far from being comprehensive and the interested reader is referred to [12, 166] for more in depth surveys. A comparison of approaches explicitly dealing with the integration of task and data parallelism can be found in [10] The Grid Computing Environments Working Group [91] maintains a list of projects related to metacomputing. In the following, we describe three native ....

D. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, 30(2):123--169, June 1998.

....substantial programming efforts [8] Low level message passing libraries also make it difficult to modify parallel programs in order to experiment with different parallel program structures. Higher level approaches to parallel application development may provide different levels of abstraction [9]. High abstraction levels such as functional languages leave many decisions related to the program behavior to the parallelization framework and therefore attain only moderate performance. Intermediate abstraction levels, such as CC [10] skeletons [11] 12] 13] and Mentat [14] leave the ....

D. Skillicom, D. Talia, Models and Languages for Parallel Computation, ACM Computing Surveys, Vol. 30, No. 2, pp. 123-169, June 1998

....can be shown with the characteristic function as discussed in Chapter 5. The use of the characteristic function is not restricted to synchronous models, but can also be implied on models with synchronous sub domains. Skillicorn and Talia discuss models of computation for parallel architectures in [105]. Their community faces similar or even identical problems as are typical for SoC design since a SoC architecture often includes a number of parallel microprocessors or other parallel hardware blocks. In fact all typical parallel computer structures (SIMD, MIMD ) can be implemented on a SoC ....

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

....the di erent concurrency properties of a system. This section present a brief survey of other concurrency models found in literature. It will also discuss the relation between these models and our model. For an extensive discussion on models and languages for parallel computation, we refer to [33]. The article surveys parallel programming models and languages and assesses their suitability for realistic portable parallel programming. They argue that an ideal model should be easy to program, should have a software development methodology, should be architecture independent, should be easy ....

Skillicorn, D.B. and D. Talia, Models and languages for parallel computation. In: ACM Computing Surveys, Volume 30, Issue 2, June 1998. New York, USA: ACM Press. p. 123-169.

....process. Ideally, parallelism should be fully implicit. This cannot be achieved for any kind of application with the available tools and techniques. Nor is it always desirable, since this would impede the experimentation with novel, more performant parallel algorithms. As identified in [97], the existing languages and systems can be classified according to the ability of making implicit the following: Parallelism Decomposition Mapping 7.4. A DISTRIBUTED ALGORITHM FOR DATA CONSISTENCY 87 Synchronization. Different approaches cover subsets of the above from all to nothing, i.e. ....

....support data parallel applications have been proposed [31] 54] Either they do not support complex data representations with general distribution, or many of the concurrency aspects are still visible to the user. A complete survey of models and languages for parallel computation can be found in [97]. We will only refer to the object oriented models. In [97] they are classified into external and internal models according to whether the parallelism is orthogonal to the object model, or integrated with the object model. We are interested in the internal object models, because these are closely ....

[Article contains additional citation context not shown here]

David B. Skillicorn and Domenico Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123,169, June 1998.

....minimal impact on the application s code (non intrusiveness) and should increase concurrency while avoiding deadlocks, i.e. provide both liveness and safety. 3 Related Work There exist numerous languages or libraries for parallel programming with various levels of transactional support (see [11] for a survey) They introduce high level tools and paradigms adapted to the development of parallel applications, by enabling the decomposition of complex programs into multiple tasks that can execute concurrently on parallel or distributed architectures. When available, transactional semantics ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123-- 169, 1998.

....memory as an abstraction over complicated interaction in implementing the reading, writing, addressing, and paging are merely handled at another level. This virtual architecture modeling method is precisely the approach taken in distributed shared memory systems (see e.g. Skillicorn et al. [602]) These systems provide an interface to the programmer as if a single physically shared memory were available. But, in reality, behind the scenes it is being simulated. This means that virtual architectures are a way of making the transition from one level to another explicit. That is, they ....

Skillicorn, D. B., and Talia, D. Models and languages for parallel computation. ACM Computing Surveys, 30(2), 1998, pp. 123--169.

....in p=16 consecutive p=16 linear p=16 extended p=32 consecutive p=32 linear p=32 extended Figure 8. Runtimes of the different execution schemes of the extrapolation method with 4 different stepsizes for # ##and # ##processors (top) and for # ###and # ###processors (bottom) of a Cray T3E. [2, 19] for an overview of systems and approaches and see [4] for a detailed investigation of the benefits of combining task and data parallel executions. Most closely related to our work concerning the parallel programming model are approaches which combine multiprocessor task and data parallelism. ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, 1998.

....to avoid the overhead of parallelism if only one processor is involved. Note that the efficiency of the parallel prefix algorithm is bounded by 50 since the parallel algorithm must perform two sweeps over the array while one sweep is sufficient for the sequential algorithm. 4 Related work [15] provides a recent systematic survey of parallel programming languages. Parallel Java extensions. There are several approaches to introducing HPF like data parallelism to Java. For instance, HPJava [4] adds distributed arrays corresponding to a subset of HPF. Naturally there are some ....

D. B. Skillicorn, D. Talia. Models and Languages for Parallel Computation. ACM Surveys, June 1998.

....programming in that there are many different languages to 15 select from to solve a problem. In this section, we restrict our attention to three dominant programming approaches data parallel, message passing, and sharedmemory. A good survey on this topic is the paper by Skillicorn and Talia [ST98]. 2.2.1 Data Parallel The data parallel model provides constructs for expressing that a statement sequence is to be executed in parallel on different data. Data parallel languages are attractive because parallelism is not expressed as a set of processors whose interactions are managed by the ....

David B. Skillicorn and Domenica Talia. "Models and Languages for Parallel Computation." ACM Computing Surveys, 30(2):123--169, June 1998.

....parfor and seqfor loop over module actications of MicroSteps annotated with the same group size. 5. RELATED WORK A large variety of programming models and languages have been proposed with different levels of abstraction. A good overview of related work on programming paradigms can be found in [9, 16]. The most important language approaches include functional programming languages and algorithmic skeletons for which parallelism is implicitly available as independent expressions can be evaluated in parallel, and data parallel languages like HPF (High Performance Fortran) 5] or NESL [3] Other ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, 1998.

....operators. Further examples are the use of programmer oriented abstractions on top of MPI in the PEMPI system at Basel University [11] and a high level extension of C , PROMOTER, developed at GMD FIRST Berlin [13] For an excellent survey of a variety of high level parallel models, see [21]. Some ongoing projects were presented at a recent Dagstuhl seminar [9] The main novelty of our FAN framework is the intensive use of program transformations in the early stages of design process, supported by corresponding cost models and programming tools. The framework is ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 1998.

.... theoretical results, empirical results, or experimental parameterization of BSP programs [1, 2, 3, 4, 5, 21] Many alternative models of parallel computation have been proposed in the literature a good survey on this topic are papers by Maggs, Matheson, and Tarjan [9] and Skillicorn and Talia [13]. Several models exist to support heterogeneous parallel computation. However, they are either primarily of theoretical interest or are basically languages runtime systems without a solid theoretical foundation. For an overview of these approaches, we refer the reader to the surveys by Siegel et ....

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

.... the problem oriented software construction from the low level concerns of the parallel execution (communication, synchronization and mapping) It is well known that more abstract models usually allow easy software construction on top of them, but are themselves hard to implement efficiently [12]. Since we do not have the ambitious goal to develop a model for generalpurpose parallelism, we will follow the patterns of the typical applications described in the introduction. Using the visualization of a CT scan (Fig. 1) as an example, we can identify the following entities which compose ....

Skillicorn D. B., Talia D.: Models and Languages for Parallel Computation. October 1996. http://www.qucis.queensu.ca/home/skill/models.ps

....scientific applications. This standard, based on other essays like X3H5, has been designed to allow an easier programming, and to be architectureindependent and efficient on shared memory architectures, which some programmers consider as the main features of an ideal parallel programming paradigm ([11]) OpenMP is a set of compiler directives, along with some library functions 2 ( 2] The compiler directives allow mainly parallel sections (and particularly the automatic loop decomposition) single sections (executed by only one thread) critical sections, and barriers. The scope attributes ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

.... problem in parallel computing is to the determine the extent to which one can write parallel programs using a high level general purpose and architecture independent programming model and have them executed on a variety of parallel and distributed architectures without sacrificing efficiency [1]. Programming models should reflect the cost of executing operations at a level that is visible to the software. The information provided by the cost model on the target machine should be as little and as abstract as possible. One of the most difficult tasks is to incorporate the effect on ....

D. B. Skillicorn and D. Talia, "Models and Languages for Parallel Computation." submitted to ACM Computing Surveys, 1996.

....which the von Neumann model has played in sequential computing. The emergency of such a model would stimulate the development of a new parallel software industry and provide a clear focus for future hardware developments. For a model to succeed in this role, it should offer some basic properties [145]. 1. It should be easy to program. A model should hide as much as possible the details of parallel execution. These include the decomposition into parallel threads, the thread mapping, the communication and the synchronization between threads. 2. It should provide a software development ....

D. B. Skillicorn and D. Talia. Models and Languages for Parallel Computation. submitted to ACM Computing Surveys, 1996.

....mechanisms is proposed on the grounds of the experiences made. 2 Structured Parallel Programming The software engineering problems we mentioned in the introduction follow from the fact that most high performance computing technologies do not fully obey to the principles of modular programming [11]. Our work on parallel programming models and languages has led us to the concept of parallel coordination language. This is the main feature of the PPE SkIE. The constructs of the language allow to express the global structure of the program, providing a high level description that is machine ....

D. B. Skillicorn, D. Talia, Models and languages for parallel computation, ACM Computing Surveys 30 (2) (1998) 123-169.

....HPRI 1999 CT 00026 (the TRACS Programme at EPCC) and scheduling [1, 4] enable the design of automatic compilation techniques as a pathway to portability and paramount optimal performance. Furthermore, SP programming is significantly easier and less error pruning than unstructured programming [15]. Several examples of SP programming are found throughout the parallel programming domain: the Bird Meertens Formalism [14] SCL [3] BSP [17] NestStep [10] SPC [5] Cilk [2] and, to some extent, OpenMP [11] 1 . Beside the obvious advantages of SP programming models, a critical question is ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

....be used for rapid prototyping and allow the exploration of alternative algorithms. Finally, the underlying implementation can exploit a wide range of optimisations (in both time and space) that are specific to the strategy concerned. Examples of the latter is illustrated in the work on skeletons [4, 8] which has resulted in several frameworks being studied and implemented in parallel. This paper concentrates on a particular class of algorithms with relevance to an important number of real life applications, namely genetic algorithms. It defines a simple framework for the definition of such ....

D. B. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, 30(2):123--169, June 1998.

....can all be utilized as components to create a distributed application and to implement a distributed shared memory protocol and other abstract programming interfaces. The survey by Skillicorn and Talia discusses the classifications of these and other models for parallel and distributed computing [ST98]. 2.2 Distributed Shared Memory The following distributed shared memory (DSM) systems address issues relating to the development of efficient distributed systems for high performance computing. DSMs attempt to hide the data distribution from the user by implementing a shared conventional memory ....

David B. Skillicorn, & Domenico Talia. (1998, June). Models and languages for parallel computation. ACM Computing Surveys, 30(2), 123-169.

....in the models like BSP [8] and LogP [9] and towards introducing higher level features as in HPF. What is lacking at the low level end is a sound methodology for designing algorithms and programs. This brief overview (for the details of particular models and languages see an excellent survey [10]) shows that the both ends of the spectrum are actually moving towards each other. Such a development raises the need, and at the same time it creates the prerequisites, for nding a common ground, on which the advantages of both high level and low level models could be combined, thus enabling the ....

D. Skillicorn and D. Talia, \Models and languages for parallel computation," ACM Computing Surveys, 1998.

....Portability, the programs should be both code and performance portable across many di erent parallel machines. Massive Parallelism, the underlying software components should implicitly exploit the available parallelism in a program to the maximum degree. Algorithmic Skeletons is a general model [153] for programming parallel machines which was introduced by Murray Cole [40] in 1989. This model provides high abstraction, portability across di erent architectures and high performance. However, algorithmic skeletons prevent the programmer from writing parallel algorithms where it is not possible ....

D.B. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, 30(2):125-169, June 1998.

....structures are restricted to the class of series parallel (SP) DAGs offer cost performance advantages with respect to both scheduling [1, 3] cost estimation [9, 5, 15, 17, 16] and, last but not least, programming itself. These attractive properties, as also mentioned by Skillicorn and Talia [18], have led to a range of parallel programming models such as BSP [20] LogP [2] and SPC [5] Despite the obvious advantages of SP programming models, however, a critical question is to what extent the ability to express parallelism is sacrificed by restricting parallelism to SP form. Note that ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

....However, in our full system, the use of a HOF will be no guarantee of parallelism: that will depend on program analysis determining that such parallelism is useful in the sense of computation outweighing communication. In their survey of models and languages for parallel computation [13], Skillicorn and Talia identify six stages between Nothing Explicit, Parallelism Implicit (NEPI) models which abstract away from all aspects of parallelism and Everything Explicit (EE) models where all aspects must be treated explicitly. They note, as have others, that exploiting the implicit ....

D. B. Skillicorn and D. Talia. Models and Languages for Parallel Computation. Computing Surveys, June 1998.

....belongs to a limited set of forms. Interest in restricted models has arisen recently, as models allowing arbitrary computations structures to be expressed (the universal models) have been proved too powerful to be implemented efficiently on current (and reasonably future) commercial architectures [24, 22, 20, 15, 16]. When developing a high level parallel programming system, we need to solve automatically problems such as mapping, scheduling, data distribution, etc. All those problems are intractable for universal models, whereas in principle can be made tractable for restricted ones. The skeleton based ....

D.B. Skillicorn and D. Talia. Models and languages for parallel computation. To appear on ACM Computing Surveys, 1998.

....It is well known that the use of SP structured parallel programming models has a number of advantages with respect to cost estimation [8, 11, 4, 14, 12] scheduling [3, 1] and last but not least, ease of programming. These attractive design properties, as outlined by Skillicorn and Talia [13], have led to a range of parallel programming models such as BSP [16] LogP [2] and SPC [4] However, expressing a problem in a parallel language that imposes SP form may imply loss of parallelism since, compared to the original NSP version, additional synchronizations may have to be added. ....

David B. Skillicorn and Domenico Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

No context found.

D.B. Skillicorn and D. Talia, "Models and languages for parallel computation", ACM Computing Survey, 30:123-169, 1998.

....focus areas for further research and development in Section 5. Conclusions are given in Section 6. 3. Properties for Grid Programming Models There are several general properties that are desirable for all programming models. Properties for parallel programming models have also been discussed [64]. Grid programming models inherit all of these. The grid environment, however, will shift the emphasis on these properties dramatically to a degree not seen before. Useability. Grid programming tools should support a wide range of programming concepts and paradigms, e.g. simple desktop ....

D. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2), June 1998.

No context found.

D. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, vol. 30, issue 2, pp 123 -- 169, 1998.

No context found.

D.B. Skillicorn, D. Talia, Models and languages for parallel computation, ACM Computing Surveys 30 (1998) 123-169.

No context found.

Skillicorn, D.B., Talia, D.: Models and languages for parallel computation. ACM Computing Surveys 30 (1998) 123--169

No context found.

D. Skillicorn and D. Talia, "Models and languages for parallel computation," ACM Computing Surveys, vol. 30, no. 2, pp. 123--169, 1998.

No context found.

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

No context found.

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

No context found.

Skillicorn, D.B., Talia, D.: Models and languages for parallel computation. ACM Computing Surveys 30 (1998) 123--169

No context found.

D.B. Skillicorn and D. Talia, Models and languages for parallel computation, ACM Computing Surveys V.20 N.2 1998.

No context found.

D.B. Skillicorn and D. Talia. "Models and languages for parallel computation". ACM Computing Surveys, 20(2):123--169, June 1998.

No context found.

D.B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 20(2):123--169, June 1998.

No context found.

D.B. Skillicorn and D. Talia. \Models and languages for parallel computation". ACM Computing Surveys, 20(2):123-169, June 1998. 2

No context found.

D.B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 20(2):123--169, June 1998.

No context found.

Skillicorn, D.B., Talia, D.: Models and languages for parallel computation. ACM Computing Surveys 30 (1998) 123--169

No context found.

David B. Skillicorn and Domenico Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123-169, June 1998.

No context found.

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

No context found.

D. Skillicorn and D. Talia. Models and Languages for Parallel Computation. ACM Computing Surveys, 30(2):pages 123-169, 1998.

No context found.

D. B. Skillicorn and D. Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

No context found.

D. Skillicorn and D. Talia. Models and languages for parallel computation, 1998.

No context found.

David B. Skillicorn and Domenico Talia. Models and languages for parallel computation. ACM Computing Surveys, 30(2):123--169, June 1998.

First 50 documents  Next 50

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