A. J. van Gemund. Performance prediction of parallel processing systems: The PAMELA methodology. In Proceedings of the 7th ACM ICS, pages 318--327, Tokyo, July 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....cannot change during its normal execution; only at rendezvous points or when the process delays can time change. A process can be aware of the centralized time, but it cannot influence the current model time except by delaying itself. The choice for modeling time was in part influenced by Pamela [30], a run time library that is used to model parallel programs. 20.3.5 Differences from Original CSP Model as Proposed by Hoare The model of computation used by the CSP domain differs from the original CSP [40] model in two ways. First, a notion of time has been added. The original proposal had no ....

....the notion of time. One solution is to use some other timed model of computation, such as DE, for describing temporal properties. Another solution is to extend the process networks model of computation with a notion time, as we have done in Ptolemy II. This extension is based on the Pamela model [30], which was originally developed for modeling the performance of parallel systems using Dykstra s semaphores. In the timed PN domain, time is global. All processes in a model share the same time, which is referred to as the current time or model time. A process can explicitly wait for time to ....

A. J. C. van Gemund, "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology," Proc. 7th Int. Conf. on Supercomputing, pages 418-327, Tokyo, July 1993.

....behavior. Mostly models (queueing networks, Petri nets, stochastic task graphs, etc. are used to describe the behavior of the program, and the resulting analysis technique is called performance prediction based on models. A few approaches of this kind have been reported in the lit erature [5, 15, 20, 23], some of which could be successfully applied in restricted application areas [24] The main point of criticism is the empirical fact that most applications fail to achieve the performance predicted based on analytical models. Furthermore, the complexity of implementation details that have ....

A. J. C. van Gemund. Performance Prediction of Parallel Processing Systems: The PAMELA Methodology. In Proc. 1993.

....of model based performance prediction, models are built explicitly with the motivation to be able to derive performance indices for a range of different parameter settings, i.e. workload or system parameters. Explicit modeling is not new as a performance engineering method for distributed systems [6,7], and could be applied successfully in areas where model assumptions are sufficiently close to system characteristics. Success of the approach is almost guaranteed when problems of static nature are investigated in homogeneous environments with deterministic operational characteristics [8] ....

A.J.C. van Gemund, Performance prediction of parallel processing systems: the pamela methodology, in: Proceedings of the 1993.

....message events are partially ordered, rather than totally ordered as they would be were they placed on a time line. Both models of computation can be augmented with a notion of time to promote interoperability and to directly model temporal properties (see for example [76] In the Pamela system [85], threads assume that time does not advance while they are active, but can advance when they stall on inputs, outputs, or explicitly indicate that time can advance. By this vehicle, additional constraints are imposed on the order of events, and determinate interoperability with timed models of ....

A. J. C. van Gemund, "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology," Proc. 7th Int. Conf. on Supercomputing, pages 418327, Tokyo, July 1993.

....process. If the algorithm graph is symmetric, an expression for communication overhead can be derived. A resource oriented approach, specifying the dependence structure and the demands for computation and communication on the corresponding resources (processors, communication network) is used in [Gemu 93b] The structure is described in a program description language. The parallel architecture is also modeled in this language. Performance prediction is based on the method of serialization analysis. Andre [Andr 87] also characterizes the program by its communication structure using a description ....

A. J. C. van Gemund. "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology". In: Proc. 1993 ACM Int. Conf. on Supercomputing, July 1993, Tokyo, Japan, ACM, 1993.

....and not reversible without significant efforts. This misconception is addressed by more recent work which tries to move performance engineering up to the front of the development cylce. Several approaches for performance prediction based on models have been reported recently [Mak 90, Sree 92, Gemu 93a] some of which could be successfully applied in restricted application areas [Wabn 94] The drawback of single parameter set model evaluation strategies is removed by very recent ideas to automate parameter variation and scalability analysis based on models [Gram 93, Malo 94] Applications in ....

A. J. C. van Gemund. "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology". In: Proc. 1993 ACM Int. Conf. on Supercomputing, July 1993, Tokyo, Japan, ACM, 1993.

....message events are partially ordered, rather than totally ordered as they would be were they placed on a time line. Both models of computation can be augmented with a notion of time to promote interoperability and to directly model temporal properties (see for example [74] In the Pamela system [83] , threads assume that time does not advance while they are active, but can advance when they stall on inputs, outputs, or explicitly indicate that time can advance. By this vehicle, additional constraints are imposed on the order of events, and determinate interoperability with timed models of ....

A. J. C. van Gemund, "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology," Proc. 7th Int. Conf. on Supercomputing, pages 418327, Tokyo, July 1993.

....running on di erent architectures, including SPDs. The simulation model outputs the execution time of the application, the utilisation of the various hardware components, and other performance measures. Our performance model is implemented in Pamela (PerformAnce ModEling LAnguage) [11]. Pamela is a C style procedure oriented simulation language in which a number of operators model the basic features of a set of concurrent processes. A detailed overview on Pamela is given in [11] 2.2 Model components Software application The hierarchical treecode [1] is one of the most ....

....performance measures. Our performance model is implemented in Pamela (PerformAnce ModEling LAnguage) 11] Pamela is a C style procedure oriented simulation language in which a number of operators model the basic features of a set of concurrent processes. A detailed overview on Pamela is given in [11]. 2.2 Model components Software application The hierarchical treecode [1] is one of the most popular numerical methods for particle simulation involving long range interactions. It is widely used in the Computational Astrophysics community to simulate systems like single galaxies or clusters of ....

van Gemund, A.: Performance prediction of parallel processing systems: The Pamela methodology. In: Proceedings of seventh ACM International Conference on Supercomputing, ACM press (1993)

.... ; RGB2YUV DCT BlockData tokensize = 64; DCT Q BlockData tokensize = 128; numbermembufs = 4; membufsize = 128; Schedulers mP : default ; RGB2YUVDCT : default ; Figure 6. Fragment of the description of the mapping shown in Figure 7 model is based on the Pamela [16] multi threading environment, where each Kahn process is executed in a separate thread. The simulation of the architecture model is currently based on TSS (Tool for System Simulation) which is a Philips in house architecture modeling and simulation framework. In order to evaluate a system, the ....

A.J.C. van Gemund, "Performance prediction of parallel processing systems: The PAMELA methodology," in Proc. 7th ACM Int. Conference on Supercomputing, Tokyo, July 1993, pp. 318--327.

....describe realistic video processing architecture instances. 6. Execution Model We simulate dataflow architectures to get performance numbers. As explained before, the simulation speed, and thus level of modeling, is an important feature of a retargetable simulator. The PAMELA work of Van Gemund [12] shows that it suffices to model two constraints in parallel architectures, namely condition synchronization of data and mutual exclusivity of resources, to get clock cycle accurate performance numbers. We can model these constraints using the concepts of processes, semaphores and delays. These ....

A. J. van Gemund. Performance Prediction of Parallel Processing Systems: The PAMELA Methodology. In Proc. 7th ACM Int. Conference on Super computing, pages 318--327, July 19-23, 1993.

....This method requires detailed knowledge of the underlying system, and is more oriented towards providing directions for efficient data distributions. Van Gemund [64] proposed a compile time mechanism to predict the performance of parallel systems. His approach was based on the Pamela formalism [63], consisting of an imperative set of constructs to explicitly express concurrent operations. As a side effect of compilation, he derived symbolic expressions representing the execution time of the set of constructs, accounting 11 for computation, communication, and overhead due to resource ....

Arjan J. C. van Gemund. Performance prediction of parallel processing systems: The PAMELA methodology. In Proceedings of the 7 th ACM International Conference on Supercomputing, pages 318--327, Tokyo, July 1993.

No context found.

A.J.C. van Gemund. Performance Prediction of Parallel Processing Systems: The PAMELA Methodology. in Proc. 7th ACM Int. Conf. on Supercomputing, pp. 318-327 Tokyo, July 1993. 40 BIBLIOGRAPHY 41

....static analysis inherently disregards the dynamic aspects typical for macro data ow execution such as processor sharing and communication link sharing. The prediction approach we propose is based on performance simulation in which we use a formalism called Pamela (PerformAnce ModEling LAnguage [1,2]) Unlike conventional simulation approaches, our approach is based on substantial model reduction, yet without signi cant loss of accuracy. The underlying motivation for our approach is that a prediction technique that combines a high prediction speed with a guaranteed minimum accuracy (within ....

....startup delay. Hence communication delay is dominated by bandwidth rather than startup delay. 3. PERFORMANCE SIMULATION 3.1. Pamela For the sake of completeness and understanding of the rest of the paper, we brie y present the main features of the language. More details can be found in, e.g. [1]. Pamela is a simple, process oriented simulation formalism. Apart from the conditional control ow operators if and while, Pamela includes binary (in x) operators to describe sequentialism (i.e. and fork join parallelism (i.e. k ) The parallel operator has implicit barrier semantics. In ....

[Article contains additional citation context not shown here]

A.J.C. van Gemund, \Performance prediction of parallel processing systems: The Pamela methodology," in Proc. 7th ACM Int. Conf. on Supercomputing, Tokyo, 1993, pp. 318-327.

....cost and symbolic properties of SP task graph analysis, however, their inherent inability to model mutual exclusion makes them generally unsuitable as the basis for a generalpurpose performance modeling technique. 2. 3 Approach Recently, a symbolic performance modeling approach has been presented [20, 21, 22, 23] that combines the low cost, symbolic properties of deterministic task graph analysis with an approximate, symbolic analysis of the effects of mutual exclusion. The approach provides a prediction accuracy that is acceptable during the first stages of parallel (embedded) system design in which ....

A.J.C. van Gemund, "Performance prediction of parallel processing systems: The Pamela methodology," in Proc. 7th ACM Int'l Conf. on Supercomputing, Tokyo, July 1993, pp. 318-- 327.

....application of serialization analysis, a low cost extension to the traditional static scheme which approximates the effects of contention. The analysis method is defined in terms of a performance modeling formalism called Pamela (PerformAnce ModEling LAnguage) introduced in last year s Conference [12]. While that paper presented the formalism and rationale, rather than a full fledged performance calculus, this paper presents a linear time algorithm yielding a lower bound which we show to be close to results obtained from simulation (which accurately accounts for contention) Thus, a novel, ....

....defined for seriesparallel structured models, in this paper we consider a more restricted mathematical formalism, which effectively defines a simple (process) algebra. In the following, we introduce only those operators which will be used in the rest of the paper. For a more detailed treatment see [12, 14]. The (restricted) Pamela formalism is organized around three temporal operations, i.e. for sequentialism, k for fork join parallelism with implicit barrier synchronization, and use(r; for mutual exclusion, where a process accesses a (binary) resource r for the duration of (virtual) time ....

[Article contains additional citation context not shown here]

A.J.C. van Gemund, "Performance prediction of parallel processing systems: The Pamela methodology," in Japan, July 1993, pp. 318--327.

....and error prone process, that is akin to manual complexity analysis of parallel algorithms [4, 14] Aimed to provide tool support in this derivation process, a performance simulation formalism called PAMELA 2 Arjan J. C. van Gemund. PerformAnce ModEling LAnguage) has been presented [7]. Both the parallel program and machine are modeled in terms of PAMELA. Instead of simulation, the PAMELA model is mechanically mapped into a symbolic cost model that has the lowest possible solution complexity, while offering a prediction accuracy that is sufficient to discriminate between ....

A.J.C. van Gemund, "Performance prediction of parallel processing systems: The PAMELA methodology," in Proc. 7th ACM Int'l Conf. on Supercomputing, Tokyo, 1993, pp. 318--327.

....in a parallel processing system with a shared memory. The latter problem has been investigated by many researchers, some of whom provide a specific solution [1, 2, 3] Methods have also been developed that provide lower bounds on response times for parallel systems with deterministic service times [4]. It is realistic to assume that a memory read write cycle has a constant value. However, the time a processor (job server) is busy between memory accesses is not constant, and has been approximated by a negative exponential distribution. A very complicated expression requiring Laplace ....

A.J.C. van Gemund, "Performance Prediction of Parallel Processing Systems: The Pamela Methodology, " Proceedings 7th ACM Int'l Conference on Supercomputing, Tokyo, July 1993, pp. 318-- 327.

.... parallel programs is an effort intensive and error prone process, that is akin to manual complexity analysis of parallel algorithms [6, 18] Aimed to provide tool support in this derivation process, a performance simulation formalism called PAMELA (PerformAnce ModEling LAnguage) has been presented [8, 9]. Both the parallel program and machine are modeled in terms of PAMELA. Instead of simulation, the PAMELA model is automatically compiled into a symbolic cost model that has the lowest possible solution complexity, while offering a prediction accuracy that is sufficient to discriminate between ....

....of programs to PAMELA models can be easily mechanized. In practice this implies that a data parallel program can be compiled into a symbolic cost model within seconds, while the cost model predicts performance within milliseconds. Building on the principles as described in previous publications [8, 9], in this paper we present an automatic symbolic cost estimator for data parallel programs. The tool has been developed as part of the JOSES project, a European funded research project aimed at developing compiler technology for embedded (multi)processor systems [10] The cost estimator is ....

[Article contains additional citation context not shown here]

A.J.C. van Gemund, "Performance prediction of parallel processing systems: The PAMELA methodology," in Proc. 7th ACM Int'l Conf. on Supercomputing, Tokyo, July 1993, pp. 318-- 327.

No context found.

A. J. van Gemund. Performance prediction of parallel processing systems: The PAMELA methodology. In Proceedings of the 7th ACM ICS, pages 318--327, Tokyo, July 1993.

No context found.

A. J. C. van Gemund. Performances prediction of parallel processing systems: The pamela methodology. In Proc. 7th ACM International Conference on Supercomputing, pages 318--327, Tokyo, Japan, July 1993.

No context found.

A. J. C. van Gemund, "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology," Proc. 7th Int. Conf. on Supercomputing, pages 418-327, Tokyo, July 1993.

No context found.

A. J. C. van Gemund, "Performance Prediction of Parallel Processing Systems: The PAMELA Methodology," Proc. 7th Int. Conf. on Supercomputing, pages 418-327, Tokyo, July 1993.

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