When implementing parallel programs for parallel computer systems the performance scalability of these programs should be tested and analyzed on different computer configurations and problem sizes. Since a complete scalability analysis is too time consuming and is limited to only existing systems, extensions of modeling approaches can be considered for analyzing the behavior of parallel programs under different problem and system scenarios. In this paper, a method for automatic scalability analysis using modeling is presented. Initially, we identify the important problems that arise when attempting to apply modeling techniques to scalability analysis. Based on this study, we define the Parallelization Description Language (PDL) that is used to describe parallel execution attributes of a generic program workload. Based on a parallelization description, stochastic models like graph models or Petri net models can be automatically generated from a generic model to analyze performance for scaled parallel systems as well as scaled input data. The complexity of the graph models produced depends significantly on the type of parallel computation described. We present several computation classes where tractable graph models can be generated and then compare the results of these automatically scaled models with their exact solutions using the PEPP modeling tool. 1

Mean value analysis (MVA) is a well-known solution technique for separable closed queueing networks used in performance modeling of computer and communication systems. In many cases, like for sensitivity analysis or with inaccurate model input parameters, intervals are more appropriate as model inputs than single values. This paper presents a version of the MVA algorithm for separable closed queueing networks with one customer class consisting of load-independent queueing centers as well as delay devices, which accepts both single values and intervals as input parameters in arbitrary combination. Monotonicity of the model outputs with respect to all input parameters is proved and these monotonicity properties are used to construct a low cost intervalversion of the MVA algorithm providing exact output intervals as results. Thus, dependency problems commonly arising with the interval evaluation of arithmetic expressions are avoided without significant increase in computation costs. Addit...

Bounding techniques for queueing network models used to analyze the performance of parallel and distributed computer systems accept single values as model inputs. Uncertainties or variabilities in service demands may exist in many types of systems. Using models with a single aggregate mean value for each parameter for such systems can lead to inaccurate or even incorrect results. This paper proposes to use histograms for characterizing model parameters that are associated with uncertainty and/or variability. The adaptation of the well-known asymptotic bounds as well as balanced job bounds for single class queueing networks to histogram parameters is presented in the paper. 1 Introduction Distributed computer systems require effective tools for predicting their performance and analyzing their behavior. Analytic models such as single class queueing networks can be used for performance estimation of such systems [15]. These techniques are popular because of their relatively low cost in ?...

In order to program paraller ' and distributed systems ef-jciently, a systematic way ojl " dejning and understanding the complex behavior of process interactions in concurrent programs is needed. A proven method for understand-ing existing programs is event-driven monitoring which abstracts the program behavior to a sequence of events. These event traces are analyzed for debugging and tuning of the program. A new method is model-driven monitoring. By building a functional model of the parallel program describing the essential properties for debugging and performance evalu-ation, the deJinition of monitoring events is automated and carried out systematically. This integration of modeling and monitoring guarantees the same set of events in mon-itoring and modeling. It enables validation, i.e. functional debugging, of the program behavior by checking the event trace against the behavior represented in the model. Ifa functional error occurs, the iillformation contained in the model allows the localization of the error in the program. However, strange performance behavior of a functionally correct program noticed during the event trace evaluation cannot easily be attached to specijic parts of the model or of the program. We solved this localization of pe~ormance errors by synchronizing event trace evaluation with event trace validation. 1.

Introduction. The Indivisibility of Functional and Temporal Behaviour Designing parallel and distributed systems we have to consider both functional specification and temporal aspects (Performance, Dependability). Usually these aspects are separated from each other. System designers use distinct hardware and software specification techniques while performance and dependability assurance is the task of modelling specialists. Such a separation was reasonable for uniprocessors and classical communication systems because it offers several advantages, mainly simplicity and understandability. Unfortunately, however, performance evaluation is often ignored and considered only when a complete misdesign is obvious. The same is true and even worse when regarding dependability aspects. The situation is changing since distributed and parallel systems became the focus of general attention and a clear trend is observable: specification techniques have been extended by time attributes and pe

Conventional performance models for computer and communication systems use single mean values as input parameters. However, uncertainties and variabilities in workload may exist in many models. For example in early stages of design only intervals may be available for the workload parameters. Furthermore, varying system behavior during different periods of time as well as the integration of dependability and performance models may lead to variabilities in parameters of the performance model. This paper proposes to use extended histograms (so-called VUlists) for characterizing model parameters that are associated with workload uncertainty and/or variability. Existing analysis techniques have to be extended to allow VU-lists as input parameters. A requirement for an algorithm to handle VU-lists is its capability to analyze interval parameterized models. Methods to adapt existing algorithms to interval parameters are discussed in the paper. Additionally, techniques which produce representa...

Abstract not found

When a parallel computation is represented in a formalism that imposes series-parallel structure on its task graph, it becomes amenable to automated analysis and scheduling. Unfortunately, its execution time will usually also increase as precedence constraints are added to ensure series-parallel structure. Bounding the slowdown ratio would allow an informed tradeoff between the benefits of a restrictive formalism and its cost in loss of performance. This dissertation deals with series-parallelising task graphs by adding precedence constraints to a task graph, to make the resulting task graph series-parallel. The weak bounded slowdown conjecture for series-parallelising task graphs is introduced. This states that the slowdown is bounded if information about the workload can be used to guide the selection of which precedence constraints to add. A theory of best series-parallelisations is developed to investigate this conjecture. Partial evidence is presented that the weak slowdown bound is likely to be 4/3, and this bound is shown to be tight.

Collecting traces with event-driven monitoring is an established and well-suited method for analyzing the dynamic behavior of parallel and distributed programs. Since these programs tend to have a very complex structure the selection of relevant events is difficult. By integrating functional modeling and event-driven monitoring, event selection is carried out in a systematic way by modeling the parallel program. This enables automation of various aspects of program analysis with monitoring, resulting in a new approach to behavior analysis called model-driven monitoring. The modeling and monitoring cycle (M 2 -cycle) has been designed to carry out model-driven monitoring efficiently. The execution of the M 2 -cycle is supported by our modeling tool PEPP. 1. Introduction When analyzing parallel and distributed programs, a systematic method is needed which helps to understand the complex dynamic behavior of concurrent processes and their interactions. The collection of event traces i...

This paper describes a teleservice - tool, which is used for the remote diagnosis of PLCS at machines. It records the operations in the interior of the PLCS. Thereby all real time - details are measured. To be able to select the measured variables flexibly, a mobile agent is constructed in the far computer. This agent will transfer into the PLCS with the help of the network and begins to work there. The observations at the software are supported by video - and audio - measurements at the machine.