H. Jonkers. Queueing Models of Parallel Applications: The Glamis Methodology. In Proceedings of the 7th International Conference on Modeling Techniques and Tools, pages 123--138, Vienna, Austria, May 1994. Lecture Notes in Computer Science, Springer Verlag.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....symbolic analysis to obtain runtime information. A. van Gemund ( 6] designed the Pamela performance modeling methodology, which provides a theoretical framework for modeling and analyzing serialization effects and the performance of parallel systems. Although petri nets ( 5] queueing networks ([10]) and markov chains ( 12] can be valuable in understanding the dynamic behavior of parallel programs, the associated analysis costs both in terms of runtime and memory requirements prohibits their use in compilers. F. Hartleb and V. Mertsiotakis ( 7] derive upper and lower execution time bounds ....

H. Jonkers. Queueing Models of Parallel Applications: The Glamis Methodology. In Proceedings of the 7th International Conference on Modeling Techniques and Tools, pages 123--138, Vienna, Austria, May 1994. Lecture Notes in Computer Science, Springer Verlag.

....of the diophantine equation ci kR = r Gamma d. For simple cases, in which c = 1; d = 0 we have r = r Gamma d = r, and ffi = R, and it follows a Gamma r b 1 Gamma r 3 PAMELA Within the variety of modeling approaches based on representations such as task graphs, queueing networks [21, 26], timed Petri nets [34] etc. the Pamela approach consists of representing program and machine in terms of an imperative formalism (called Pamela) In contrast to most simulation formalisms, however, Pamela features a performance calculus which permits (compile time) model reduction, prior to ....

H. Jonkers, "Queueing models of parallel applications: The Glamis methodology," in 7th Int. Conf. on Modelling Techniques and Tools for Comp. Perf. Eval., Vienna, May 1994.

....spent in designing and implementing the monitor base and a dozen monitor tools. 6 Analytical Modeling There are various performance modeling formalisms that can be used in modeling parallel systems. These formalisms can be classified into two classes: deterministic and probabilistic formalisms [Jonkers94] In deterministic models, all quantities are fixed. In probabilistic models, some degree of uncertainty exists, and stochastic quantities are included in the models. Queueing networks and Petri nets, which will be discussed in this section, belong to the latter class. Before reviewing the two ....

....that the number of arrivals to a device equals the number of departures from the device. 6.1. 4 Analysis of Queueing Networks In this subsection, we consider convolution algorithm and mean value analysis (MVA) They are the most widely used product form solution methods for queueing networks [Jonkers94] Gordon and Newell [Gordon67] showed that the product form solution for the distribution of the number of customers at the service centers can be computed as P n n n D D D G N M n n M n M ( 1 2 1 2 1 2 = # where D i is the total service demand per job for device i, n ....

[Article contains additional citation context not shown here]

Jonkers, H., "Queueing models of parallel applications: the Glamis methodology," in: Haring, G. and Kotsis, G. (eds.), Computer Performance Evaluation Modeling Techniques and Tools (Proceedings of the 7 th International Conference, Vienna, Austria), pp. 123-138, Springer-Verlag, May 1994.

....such as system throughput, response times, and queue lengths. There are many software tools for the analysis of queueing network models, often relying on MVA, e.g. 3] 13] Additionally, QNMs are often combined with other computer modeling and analysis techniques as proposed e.g. in [22] [10], 11] In computer systems performance analysis, there are usually various parameters (e.g. processor speed, communication rates, or workload characteristics) that have to be supplied by the analyst to evaluate the model. Uncertainties may be associated with such model parameters for various ....

H. Jonkers. Queueing Models of Parallel Applications: The GLAMIS Methodology. In Proc. of the 7 th Int. Conference on Modelling Techniques and Tools for Computer Evaluation, Vienna, Austria, May 1994.

.... MVA, e.g. 27] 6] 34] 1] There exist many software tools for the analysis of queueing network models, often with the use of MVA algorithms, e.g. 4] 5] 15] and often queueing network models are combined with other computer modelling and analysis techniques, like e.g. in [21] [13], 14] In computer systems performance analysis, there are usually various parameters (e.g. processor speed, communication speed, workload characteristics) that have to be supplied to evaluate the model. There are several reasons to be interested to describe these input parameters not as exact ....

Henk Jonkers, "Queueing models of parallel applications: The glamis methodology", in Proceedings of the 7th International Conference on Modelling Techniques and Tools for Computer Evaluation, Vienna, AUSTRIA, May 1994.

....= r Gamma d. For simple cases, in which c = 1; d = 0 we have r = r Gamma d = r, and ffi = R, and it follows ff r = d a Gamma r R e fi r = d b 1 Gamma r R e Gamma 1 2 3 PAMELA Within the variety of modeling approaches based on representations such as task graphs, queueing networks [21, 26], timed Petri nets [34] etc. the Pamela approach consists of representing program and machine in terms of an imperative formalism (called Pamela) In contrast to most simulation formalisms, however, Pamela features a performance calculus which permits (compile time) model reduction, prior to ....

H. Jonkers, "Queueing models of parallel applications: The Glamis methodology," in 7th Int. Conf. on Modelling Techniques and Tools for Comp. Perf. Eval., Vienna, May 1994.

....using pre measured kernel codes. They achieve good results for the loosely synchronous programming model. K.Y. Wang ( 10] characterizes parallel programs by a parameterized performance model, which uses symbolic analysis to obtain runtime information. Although petri nets ( 6] queueing networks ([9]) and markov chains ( 12] can be valuable in understanding the dynamic behavior of parallel programs, the associated analysis costs both in terms of runtime and memory requirements prohibits their use in compilers. F. Hartleb and V. Mertsiotakis ( 7] derive upper and lower execution time bounds ....

H. Jonkers. Queueing Models of Parallel Applications: The Glamis Methodology. In Proceedings of the 7th International Conference on Modelling Techniques and Tools, pages 123--138, Vienna, Austria, May 1994. Lecture Notes in Computer Science, Springer Verlag.

....scheme for each parallel program part, and various aspects of a program s runtime behavior. A third problem lies in producing scaled models from generic 1 Note, this generator approach could be undertaken with different modelingtechniques, including Petri net models[19] andQueuingModels [10]. representations, since model complexity, tractability, and solution accuracy must be considered. The key is to find model generation methods which produce approximately accurate models of scaled performance behavior, but that do not exceed the solution capabilities of stochastic modeling ....

H. Jonkers. Queueing Models of Parallel Applications: The Glamis Methodology. Proc. of the 7th Int. Conf. on Model. Techn. and Tools for Comp. Perf. Eval., 1994.

....on a distributed memory machine. 1 Introduction In the performance prediction of parallel systems many approaches exist which represent a specific trade off between analysis accuracy and cost. Approaches aimed for accuracy involve the use of probabilistic techniques based on queueing networks [10, 15], stochastic graphs [13, 18] timed Petri nets [1, 20] and stochastic process algebras [8] as well as simulation [16, 19] Due to the exponential state space complexity associated with these models, however, the computational costs may easily prohibit frequent use of such techniques in a design ....

H. Jonkers, "Queueing models of parallel applications: The Glamis methodology, " in LNCS, vol. 794, 1994.

....estimation approach using symbolic analysis. A. van Gemund ( 17] designed the Pamela performance modeling methodology, which provides a theoretical framework for modeling and analyzing serialization effects and the performance of parallel systems. Although petri nets ( 2, 15] queueing networks ([22]) and markov chains ( 30] can be valuable in understanding the dynamic behavior of parallel programs, the associated analysis costs both in terms of runtime and memory requirements prohibits their use in compilers. N. Yazici Pekergin and J.M. Vincent ( 33] obtain stochastic bounds on execution ....

H. Jonkers. Queueing Models of Parallel Applications: The Glamis Methodology. In Proceedings of the 7th International Conference on Modelling Techniques and Tools, pages 123--138, Vienna, Austria, May 1994. Lecture Notes in Computer Science, Springer Verlag.

....formalism and analysis algorithms to predict the completion time of parallel programs, making use of queueing networks to model the influence of the underlying parallel machine. While previous papers introduced the methodology and described algorithms to analyse a subclass of parallel programs [9, 10], this paper describes a new algorithm for the analysis of programs with arbitrary synchronisation patterns, thus generalising over the previous algorithms. It will be shown that a similar approach can be followed to include the impact of course grain, program level mutual exclusion in the ....

....of subsets of T . In combination with an instruction vector i, the function C can be specified compactly as a jT j Theta jI j matrix [C] Visit counts can then be derived by a simple matrix matrix multiplication: V ] C] Delta [F ] 5 Model analysis With the first introduction of Glamis [9], an iterative algorithm was presented for the analysis of programs with an SPS structure, i.e. programs consisting of a sequence of parallel sections, each section possibly containing different types (or classes) of tasks. In other words, the only condition synchronisations considered were ....

[Article contains additional citation context not shown here]

H. Jonkers, "Queueing models of parallel applications: The Glamis methodology," in Computer Performance Evaluation: Modelling Techniques and Tools (LNCS 794) (G. Haring and G. Kotsis, eds.), Springer-Verlag, May 1994, pp. 123--138.

....disregarded (e.g. resource contentions) On the other extreme of the trade off, in unrestricted timed Petri net models [2, 15] all important machine and program characteristics can be expressed, but the analysis cost is in general exponential to the size of the net. With the introduction of Glamis [11], a performance modelling methodology with associated analysis techniques based on an extension of queueing networks, we have chosen a position on the trade off combining accurate predictions with acceptable analysis costs. Symmetries (replications) in parallel machines and programs are exploited ....

....due to the barrier synchronisation effect. This would lead to an under estimation of the completion time, even when the maximum of the response times for the different job classes is used. In this section two solutions to this problem are presented. The first algorithm has been presented in [11] while the second algorithm is new. Although both algorithms approach the problem in a totally different manner, the results are in most cases surprisingly similar. In the cases where differences were found, the second algorithm gave more accurate results. Moreover, the second algorithm is more ....

[Article contains additional citation context not shown here]

H. Jonkers, "Queueing models of parallel applications: The Glamis methodology," in Comp. Perf. Eval.: Modelling Techniques & Tools (G. Haring and G. Kotsis, eds.), Springer, May 1994, pp. 123--138.

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