This paper presents a new domain decomposition approach whose main goal is the computation of a load balancing partition while reducing the overhead to compute such a partition. In the proposed approach, the number of neighbours of each sub-domain produced by the decomposition can be bounded by an user supplied value. This reduces the communication overhead of the appli- cation. We describe an algorithm implementing our decomposition strategy and apply our approach to WaTOR, a classical dynamical simulation problem. We report also some preliminaries result to prove the effectiveness of our approach. Domain Decomposition is a technique exploited both to balance the load and to reduce communications in parallel applications (2). It is applied when the data set of the ap- plication can be regarded as a physical domain and the computation performed on each element D of the domain requires the knowledge of a small subset of data close to D only. Several parallel applications belongs to this class, a classical example being that of cellular automata (10). This technique partitions the domain into a set of sub-domains with the same com- putational load and assigns each sub-domain to a distinct process. Each process updates the elements belonging to its sub-domain S and communicates with the other processes only to update the elements located on the boundary of S. The domain decomposition problem becomes challenging when dynamical applica- tions are considered because in these applications the decomposition of the domain has to be updated during the computation. The update is required to take into account that the number of the elements of the domain and/or their positions change dynamically during the computation. Several approaches have been proposed in the last years. Each strategy takes into ac- count the trade-off between the overhead introduced by the dynamic partitioning of the domain and the benefits obtained by balancing the load. Orthogonal Recursive Bisec- tion (6) is a domain decomposition strategy exploited in many parallel application. This technique produces an optimal balance of the work, at the expense of a large computa- tional complexity. On the other side, simpler solutions often results in decompositions of the domain characterized by an unsatisfactory load balance. This work describes the BoundedNeighbours approach, a domain decomposition technique whose main goal is to reduce the overhead introduced by partitioning the
This paper presents the analysis of relationships between relative water levels on Drava and Danube river and water level inside the Nature park "Kopacki rit" during the summer 2002. Due to high complexity of complete system under study, usual quantitative techniques couldn't be applied. Qualitative-quantitative analysis has set up some basic relationships and showed possibilities in dealing with complex water flooding systems. For the modelling procedure AI identification tool was used, based on circular qualitative-quantitative algebra.
Complex industrial processes like Marine Diesel Propulsion Plant (MDPP) have complex interrelations and interdependencies between variables and parameters. This characteristic could be used in estimating unknown or unmeasured variables from the information gathered by other measurements and sources using information fusion by means of a soft computing methods. In the paper, a structural analysis approach to identifying most relevant variables interrelations, components or subsystems of MDPP with inherent redundant information has been proposed. Sensor information fusion method was chosen to be using artificial neural networks (ANN). The paper presents proposed ANN with structure and learning algorithms. Simulation have been carried out in Matlab-Simulink environment wit engine speed estimation example.
This paper discuss an algorithm for converting plant structures generated with L-systems into the MTG format and an implementation of this method in a modeling program. The virtual laboratory software is used to integrate both a simulating program and an analyzing program into a cohesive system for modeling and analyzing plant architecture. We present sample applications of this system and focus on the assessment of virtual plants.
Recent progress has been made in simulation of off-line programmed robot motions by encapsulating original algorithms of the real control system like inverse kinematics and trajectory generation. The visualized robot motions in that kind of CAE software are realistic in the sense that the programmed desired motion of the robot is shown in real time and in a realistic way. hi the next step of the development of realistic simulation tools the user should have information about the dynamic behaviour of the robot to predict the difference between desired and actual positions. To calculate this deviations in real time, approximate dynamic models are needed which decrease decisively the effort to calculate the model and which have low loss in model accuracy. In. this paper a method to generate such efficient models is presented. The fundamental principle is to consider the motion state of the robot arm as probabilistic events. Some mathematical terms are then neglected in respect to the whole motion state. These reductions results a model, which has extensive savings of mathematical operations and has minimal effects (deviations) on the performance compare to complete model.
Petri networks and Queueing networks are two abstractions used to represent and analyze large-scale distributed systems. Though their basic ideas and objectives are different, there arc areas where a combination of these two formalisms brings new ideas and methods of both qualitative and quantitative nature. The principal problem is the analysis of discrete distributed systems with delays, services and queues that may be formally represented as queueing networks. Mathematical models of queueing networks are based on very strong, often unrealistic assumptions. That's why simulation is mostly the only feasible quantitative analytical method. The paper shows how to use Time Petri nets together with code in a high-level language to create simulation models of queueing networks. The method is especially suitable for education purposes because the only requirements are intermediate programming skills and basic knowledge of Petri nets.
The capacity of Global System for Mobile Communication (GSM) is limited and currently running into its physical limits whilst third generation mobile networks are not expected to be operational in the near future. It is, therefore, extremely important to devise new modelling and evaluation methods to enhance GSM performance for both mobile users and network operators. In this paper a novel and efficient analytic framework is devised for the performance modelling and evaluation of a wireless GSM cell based on the re-use partitioning (RUP) principle. RUP splits the available bandwidth into multiple layers of frequencies and allows a tighter frequency re-use in order to achieve a higher network capacity. A queueing network model (QNM) of a wireless GSM cell is proposed consisting of a hierarchical layer configuration which is decomposed into individual GE/GE/c/c loss models each of which is analysed in isolation via the principle of maximum entropy (ME) subject to stochastic loss system constraints. A comparative study is included to validate the relative accuracy of the derived ME closed-form expressions against simulation and assess the optimal reuse partitioning policy of the available bandwidth. Moreover, an investigation is undertaken with regard to the effect of external GSM bursty traffic upon the performance of the wireless cell.
Nowadays there exists an explosive demand of Internet web services. As a consequence, the traditional web server architecture based on a single machine (monoprocessor or multiprocessor system) is usually not sufficient to handle this increasing load because the single server upgrading process is complex, also becoming a single point of failure. Clusters of web servers, namely web clusters, connected by a fast local area network, arc emerging as an alternative for building highly scalable and high available Internet web services. This type of architecture is more scalable, more reliable and cost-effective than a single server. However, there are a lot of challenges that must be addressed to tune and increase the performance of this kind of systems. More precisely, the way workload is distributed among servers in the cluster is a crucial factor that affects global performance.Performance evaluation may be based on analytical modeling or simulation modeling. Each of them differs in their scope and applicability. However, the simulation modeling technique offers more freedom and flexibility than the other methods. In order to predict the performance of web clusters, we have implemented a very flexible and easy to use simulator that takes into account several configuration parameters. This tool can be used for both academical and research purposes. In this paper, the authors present a brief description of the employed simulation language, the queueing model in which the simulator is based on, the main input parameters and output variables, and some aspects about the internal design of the implemented tool. Finally, some experimental results obtained by using this tool are shown.
Computer simulation in physiology is an interdisciplinary problem. It is very important to provide good readability of used models. Programmer understands the architecture of models and the algorithms of programs. On the other hand physiologist usually can judge how closely the behavio of system correspond to the nature. In this text we will show how some known programming methods can help us find common language between programmers and physicians. Simulation models Simulation models in physiology are based on differential equations, regulation mechanisms and often on some mathematical algorithms like finding value of inverse function or solving some optimization problem. In scientific literature there are many articles about partial simulation models programmed in classical sequential languages like FORTRAN, MODULA, ADA, etc. Those products are hardly understandable for physiologists and non-programmers. Speak nothing of linear structure of these programs, which usually excludes any updating, adaptations or expansion. For these reasons our team found another way. We use MATLAB & SIMULINK graphical programming environment. This software is developed for use in control industry and for using advanced mathematical methods and has a strong toolset for constructing, developing and testing of various models.
The purpose. of this paper is to present the results concerning the ergodic behaviour of the deviation of the number of customers in network's model with infinite server queues. Ergodic theorems for these characteristic were proved.
The paper describes the stages of the general structure modeling process, using an object-oriented paradigm. The first part describes basic ideas about the structure modeling and implementation of the simulation mechanism in Cosmos tool. The tool was developed primarily as the basis for handling of network components and network, but is easily applied to any system type. Examples of the addition of new features to base classes, e.g. analytic and optimization procedures, and availability model are presented. The last part explains the use of developed components for IP over WDM network modeling, which had been the prime issue that initiated development of the described models.
This paper proposes the challenges introduced by the development of a simulation based DSS (Decision Support System) for maintenance of complex systems such as helicopters, trains, power plants. The paper proposes the simulation architecture devoted to face this level of complexity and the critical issue analysis related to its development.
Simulation of container terminal operation can be oriented on various important tasks. One of them can be design of the terminal and its management system. Another one can be evaluation of its functioning with possible advises for its improvement. The simulation system for container terminal should recognise extremal conditions of work and existence of bottlenecks, which should be monitored. System should be user-friendly and enable visualisation of results.
In the present paper, the periodic BMAP/M-t/c queue is analyzed. This queue. has a periodic BMAP arrival process, which is defined in this paper, and exponential service time distributions with service rates varying periodically in time. The transient distribution and several formulae for its efficient computation are derived. Furthermore, a stability condition in terms of arrival and service rates is proven, and for the case of stability the asymptotic distribution is given explicitly. A special class of periodic BMAP/M-t/c queues, which allows to reduce the complexity of computations to the level of homogeneous queues, is separated. The obtained results are applied to the spatial SMAP/M-t/(c)/(c) variant of the queue, too. This variant is very useful as a model for mobile communication networks.
A study of current server architecture models identified limitations and performance bottlenecks in servers that utilize either a pre-fork or a fork-on-demand strategy. Two new server architecture models were developed to address these problems - Automatic Rate Control (ARC) and Server Monitoring and Adaptive Response to Traffic (SMART). ARC measures the rate of incoming requests and performs an on-the-fly prediction of subsequent loads. SMART was derived from strategic management which used data collected passively to determine a future expected traffic profile and hence perform an early prediction of subsequent loads. These two models performed 40-50% better than conventional pre-fork models and were implemented as the architecture of the Broker server used as part of the application layer of the Digital Media Repository (DMR).
Computational requirements of present day applications have prompted a need for more powerful computers. A relatively cost-efficient system can be built by harvesting the powers of individual workstations. Networks (clusters) of workstations have become a popular method of achieving performance results that are very close or even beyond to proprietary supercomputers. The performance of a network of workstations can be further improved by equipping the workstations with multiple processors. Such a system is referred to as a network of symmetric multiprocessor (SMP) workstations. The programming paradigms designed for networks of uniprocessor workstations are applicable to networks of SMP workstations. However, applying a programming paradigm developed for a network of uniprocessor workstations on a network of SNIP workstations causes some unnecessary overhead. This paper presents an improved method of parallel computation on a network of SNIP workstations. The method comprises of a library called MPIT, and an MPI-like interface to the library. The library utilizes MPI to facilitate communication between the workstations, and POSIX threads to provide the parallel computation environment in a workstation. The library allows the computation and communication to overlap by creating worker threads and a communication thread. The worker threads handle the data processing while the communication thread perform all communication operations. Results from communication intensive applications show that the MPIT library provides increased performance due to the optimized communication scheme within a workstation, and due to the overlap of communication and computation.
European Space Agency Simulation Language (ESL) is a simulation language used to simulate real life continuous control systems. With the emergence of the Internet as the mainstream technology, there is ample of motivation to develop a generic architecture which allows users to conduct simulations through the Internet.In this paper a three-tier system called Distributed Simulation Management Architecture (DSMA) is presented which manages ESL simulation processes in a heterogeneous High Performance Computer (HPC) network.Analysis has shown that it is possible to accurately predict the execution time of certain type of ESL programs. An Estimation Engine is presented in this paper which is capable of estimating the execution time of this type of ESL programs with up to 94% accuracy. A scheduling algorithm called Predictive Multi-Level Priority (PMLP) scheduling algorithm is also presented here. It can effectively reduce the latency of the scheduling process of DSMA.
In this paper, the Petri net methodology was applied to model and simulate metabolic networks in the cell. An example was presented to illustrate the feasibility and potentiality of the proposed method. Provided with the rapid accumulation of metabolic data available in Internet and the development of computer techniques, it was possible to construct Petri not models automatically from biological databases. Furthermore, a proposal for biology Petri net markup language was introduced to provide a starting point for the development of a standard interchange format for bioinformatics and Petri nets.
