Alexander Quinte

Solving scheduling problems in grid resource management using an evolutionary algorithm

Evolutionary Algorithms (EA) are well suited for solving optimisation problems, especially NP-complete problems This paper presents the application of the Evolutionary Algorithm GLEAM (General Learning and Evolutionary Algorithm and Method) in the field of grid computing Here, grid resources like computing power, software, or storage have to be allocated to jobs that are running in heterogeneous computing environments The problem is similar to industrial resource scheduling, but has additional characteristics like co-scheduling and high dynamics within the resource pool and the set of requesting jobs The paper describes the deployment of GLEAM in the global optimising grid resource broker GORBA (Global Optimising Resource Broker and Allocator) and the first promising results in a grid simulation environment.

First Evaluation Results Using the New Electrical Data Recorder for Power Grid Analysis

Today, power systems are subject to fundamental changes concerning functionality and dynamics due to new energy sources and increasing demand. However, detailed information of the system and the system state is the property of the suppliers or distributors, and is not comprehensively available to researchers at present. We propose to capture easily accessible, high-rate, low-voltage (LV) time series at different locations, and to store the whole data for subsequent usage in a large database. A system state simulation shall use these data and provide for load information without knowing the currents. For this purpose, we develop the electrical data recorder (EDR) and perform measurements in our first starting test site, the island network like Karlsruhe Institute of Technology campus. We first present comparison results between captured voltage characteristics and campus smart meter measurements that we use as an indication of the load status. We develop the Electrical Grid Analysis Simulation Modeling and Visualization Tool (eASiMoV) and show feasibility in a simple simulation. We measure the storage transfer to the large-scale database facility and give evaluation results. The EDR device, the eASiMoV software, and data handling methods are exhibited as valuable components of a promising new approach.

Fast Multi-objective Scheduling of Jobs to Constrained Resources Using a Hybrid Evolutionary Algorithm

The problem tackled here combines three properties of scheduling tasks, each of which makes the basic task more challenging: job scheduling with precedence rules, co-allocation of restricted resources of different performances and costs, and a multi-objective fitness function. As the algorithm must come up with results within a few minutes runtime, EA techniques must be tuned to this limitation. The paper describes how this was achieved and compares the results with a common scheduling algorithm, the Giffler-Thompson procedure.

Introducing the Electrical Data Recorder as a new capturing device for power grid analysis

Power systems are facing fundamental changes concerning functionality and dynamics today. However, detailed data of the system are property of the suppliers or distributors and are not easily available to researchers at present. In order to understand the system dependencies we propose to derive state information by combining high-rate low-voltage time series captures at different locations together with a simulation model of the grid. We take the island network like KIT campus as our starting investigation site. In a first step we developed the Electrical Data Recorder (EDR), which is capable of recording three phase voltage time series at up to 25 kHz synchronously. All data are stored in a large scale database facility (LSDF) for subsequent usage. We intend to derive a simulation model from the comparison of the voltage characteristics to periodic smart meter measurements as the indication of the load status. In this article we introduce the new recording device and present first results.

Fast Rescheduling of Multiple Workflows to Constrained Heterogeneous Resources Using Multi-Criteria Memetic Computing

This paper is motivated by, but not limited to, the task of scheduling jobs organized in workflows to a computational grid. Due to the dynamic nature of grid computing, more or less permanent replanning is required so that only very limited time is available to come up with a revised plan. To meet the requirements of both users and resource owners, a multi-objective optimization comprising execution time and costs is needed. This paper summarizes our work over the last six years in this field, and reports new results obtained by the combination of heuristics and evolutionary search in an adaptive Memetic Algorithm. We will show how different heuristics contribute to solving varying replanning scenarios and investigate the question of the maximum manageable work load for a grid of growing size starting with a load of 200 jobs and 20 resources up to 7000 jobs and 700 resources. Furthermore, the effect of four different local searchers incorporated into the evolutionary search is studied. We will also report briefly on approaches that failed within the short time frame given for planning.

An Evolutionary Algorithm for Design Optimization of Microsystems

The concept of an evolutionary algorithm for improving the design optimization process is presented. A first application, the design optimization of a micropump, is used for both, the description of the concept being especially tailored for multicriteria applications and the presentation of very promising results.

Tackling the grid job planning and resource allocation problem using a hybrid evolutionary algorithm

This paper presents results of new experiments with the Global Optimising Resource Broker and Allocator GORBA for grid systems. The scheduling algorithm is based on the Evolutionary Algorithm GLEAM (General Learning Evolutionary Algorithm and Method) and several heuristics. The task of planning grid resource allocation is compared to pure NP-complete job shop scheduling and it is shown in which way it is of greater complexity. Two different gene models and two repair methods are described in detail and assessed by the experimental results. Based on the analysis of the experimental results, directions of further work and improvements will be outlined.

Schnelles Scheduling mit Hilfe eines hybriden Evolutionären Algorithmus (Fast Scheduling by Means of a Hybrid Evolutionary Algorithm).

Zusammenfassung Scheduling und Ressourcenplanung sind Aufgaben, die in vielen Einsatzbereichen wie z.B. der Maschinenbelegung oder Projektplanung vorkommen. In diesem Beitrag wird eine verwandte Aufgabenstellung vorgestellt, nämlich die Ressourcen- und Jobplanung im Bereich des Grid-Computing. Auf Grund der Dynamik in einem solchen heterogenen Rechnernetz ist eine schnelle Reaktion und häufige Umplanung innerhalb weniger Minuten erforderlich. Daher kommt es vor allem auf gute Resultate innerhalb eines vorgegebenen Zeitrahmens an und weniger darauf, das Optimum zu finden. Zur Lösung dieses Problems werden Heuristiken und der Evolutionäre Algorithmus GLEAM eingesetzt. Der Beitrag vergleicht mit Hilfe geeigneter Benchmarks zwei unterschiedliche Genmodelle, zwei Verfahren zur Reparatur illegaler Schedules und den Effekt eines Hilfskriteriums zur schnelleren Verbesserung von Schedules. Außerdem wird die Wirkung zweier bekannter Crossover-Operatoren für kombinatorische Optimierung untersucht. Dabei zeigt es sich, dass der Evolutionäre Algorithmus trotz der zeitlichen Einschränkung eine deutliche Verbesserung der heuristisch erzeugten Schedules erreicht.

Classification of Benchmarks for the Evaluation of Grid Resource Planning Algorithms

The present contribution will focus on the systematic construction of benchmarks used for the evaluation of resource planning systems. Two characteristics for assessing the complexity of the benchmarks were developed. These benchmarks were used to evaluate the resource management system GORBA and the optimization strategies for resource planning applied in this system. At first, major aspects of GORBA, in particular two-step resource planning, will be described briefly, before the different classes of benchmarks will be defined. With the help of these benchmarks, GORBA was evaluated. The evaluation results will be presented and conclusions drawn. The contribution shall be completed by an outlook on further activities.