Richard Olejnik

A new estimation method for distributed Java object activity

We introduce a new method to estimate the Java object activity in a distributed context of irregular applications. This method uses an observation mechanism which is itself a part of a more global load balancing system. It predicts the tendencies of the communication between the objects of the distributed application. To illustrate the behaviour of this mechanism we applied it to the Travelling Salesman Problem TSP which is solved by means of a genetic algorithm. Finally we present the overhead measurements we have done.

Byte-code scheduling of Java programs with branches for desktop grid

A method for an introductory optimization of multithreaded Java programs for execution on clusters of Java Virtual Machines (JVMs) inside desktop grids is presented. It is composed of two stages. In the first stage, a clustering algorithm is applied to extended macro data flow graphs generated on the basis of the byte-code compiled for multithreaded Java programs. These graphs account for data and control dependencies in programs including conditional branch instructions annotated by branch statistics driven from execution traces for representative sets of data. In the second stage, a list scheduling is performed based on the Earliest Task First (ETF) heuristics in which node mapping on JVMs accounts for mutually exclusive paths outgoing from conditional branch nodes. The presented object placement optimization algorithm is a part of the DG-ADAJ environment.

An observation mechanism of distributed objects in Java

We present an observation mechanism of distributed objects in the context of irregular applications developed in distributed Java. This mechanism predicts the tendencies of the communication between these objects. To ensure a good effectiveness of the execution, the obtained predictions are integrated into a distribution mechanism for the objects of the application.

Combining software agents and grid middleware

Recently, the Desktop-Grid ADaptive Application in Java (DG-ADAJ) project has been unveiled. Its goal is to provide an environment which facilitates adaptive control of distributed applications written in Java for the Grid or the Desktop Grid. However, in its current state it can be used only in closed environments (e.g. within a single laboratory), as it lacks features that would make it ready for an open Grid. The aim of this paper is to show how the DG-ADAJ can be augmented by usage of software agents and ontologies to make it more robust.

A heuristic to minimize the cardinality of a real-time task set by automated task clustering

We propose in this paper a method to automatically map functionalities (blocks of code corresponding to high-level features) with real-time constraints to tasks (or threads). We aim at reducing the number of tasks functions are mapped to, while preserving the schedulability of the initial system. We consider independent tasks running on a single processor. Our approach has been applied with fixed-task or fixed-job priorities assigned in a Deadline Monotonic (DM) or a Earliest Deadline First (EDF) manner.

Java programs optimization based on the most–often–used–paths approach

The paper presents a Java byte–code optimization algorithm, which determines an initial distribution of objects among virtual machines (JVMs) so as to decrease direct inter–object communication and balance loads of the virtual machines. The proposed optimization algorithm is based on a graph representation of control and data dependencies between methods in Java programs. These dependencies, expressed in the form of conditional macro–dataflow graphs, are discovered by a static analysis of program byte–code. n nObject placement optimization is performed before a Java program is executed in a parallel system. The optimization methods are based on the Dominant Sequence Clustering (DSC) approach. First, macro nodes are clustered on an unlimited number of processors (logical JVMs) to reduce the total program execution time. Next, clusters are merged and scheduled to adjust the number of logical JVMs to the number of real processors. The presented approach is supported by branch optimization techniques, which include detection of mutually–exclusive paths and scheduling of most–often–used–paths based on branch probabilities.

Parallel extremal optimization in processor load balancing for distributed applications

Graphical abstractDisplay Omitted The paper concerns parallel methods for extremal optimization (EO) applied in processor load balancing in execution of distributed programs. In these methods EO algorithms detect an optimized strategy of tasks migration leading to reduction of program execution time. We use an improved EO algorithm with guided state changes (EO-GS) that provides parallel search for next solution state during solution improvement based on some knowledge of the problem. The search is based on two-step stochastic selection using two fitness functions which account for computation and communication assessment of migration targets. Based on the improved EO-GS approach we propose and evaluate several versions of the parallelization methods of EO algorithms in the context of processor load balancing. Some of them use the crossover operation known in genetic algorithms. The quality of the proposed algorithms is evaluated by experiments with simulated load balancing in execution of distributed programs represented as macro data flow graphs. Load balancing based on so parallelized improved EO provides better convergence of the algorithm, smaller number of task migrations to be done and reduced execution time of applications.

Multi-objective parallel extremal optimization in processor load balancing for distributed programs

The paper concerns multi-objective methodology applied to parallel Extremal Optimization (EO) used in processor load balancing in execution of distributed programs. When load imbalance is detected in executive processors then EO algorithms are used to find best tasks migration leading to imbalance reduction and improvement of program execution time. For this a special multi-objective version of parallel EO is applied. It is based on the EO Guided Search (EO-GS) approach which employs problem knowledge to search for the best next solution state in solution improvement. In this EO version, additional fitness function is used in stochastic selection of next solution state based on computation and communication assessment of task migration targets. In the multi-objective EO approach we jointly control three objectives relevant in processor load balancing for distributed applications. They are: computational load balance in execution of distributed applications, volume of communication between tasks on different processors and task migration parameters which fight imbalance of processor loads. The proposed algorithms are assessed by simulated execution of distributed programs macro data flow graphs.

Effective processor load balancing using multi-objective parallel extremal optimization

The paper presents how Extremal Optimization can be used in a parallel multi-objective load balancing algorithm applied in execution of distributed programs. Extremal Optimization is used to find task migration which dynamically improves processor load balance in a distributed system. In the proposed multi-objective approach we use three objectives relevant to distributed processor load balancing in execution of program tasks. They are: computational load balance of processors, the volume of inter-processor communication and task migration metrics. In the algorithms additional criteria are used which are based on some knowledge on the influence of the computational and communication loads on task execution. The proposed algorithms are assessed by simulation experiments with distributed execution of program macro data flow graphs. Two methods of finding compromise solutions based on the Pareto front were used: one based on a geometric (Euclidean) distance of solutions and the second one based on the Manhattan (taxicab geometry) distance. The influence of the distance geometry on the final solutions is discussed.

Multi-Objective Extremal Optimization in Processor Load Balancing for Distributed Programs

The paper presents a multi-objective load balancing algorithm based on Extremal Optimization in execution of distributed programs. The Extremal Optimization aims in defining task migration as a means for improving balance in loading executive processors with program tasks. In the proposed multi-objective approach three objectives relevant in processor load balancing for distributed applications are jointly optimized. These objectives include: balance in computational load of distributed processors, total volume of inter-processor communication between tasks and task migration metrics. In the proposed Extremal Optimization algorithms a special approach called Guided Search is applied in selection of a new partial solution to be improved. It is supported by some knowledge of the problem in terms of computational and communication loads influenced by task migration. The proposed algorithms are assessed by simulation experiments with distributed execution of program macro data flow graphs.