Ana Cortés

A distributed diffusion method for dynamic load balancing on parallel computers

Parallel applications can be divided into tasks that can be executed simultaneously in different processors. Depending on prior knowledge about computational requirements of the problem, the assignment of tasks to processors can be guided in two ways: static and dynamic. We propose a new dynamic load balancing algorithm based on the diffusion approach which employs overlapping balancing domains to achieve global balancing. Since current diffusion methods consider discrete units, the algorithms may produce solutions which, although they are locally balanced prove to be globally unbalanced. Our method solves this problem taking into account the load maximum difference between two processors within each domain, providing a more efficient load balancing process.<<ETX>>

Applying a Dynamic Data Driven Genetic Algorithm to Improve Forest Fire Spread Prediction

This work represents the first step toward a DDDAS for Wildland Fire Prediction where our main efforts are oriented to take advantage of the computing power provided by High Performance Computing systems to, on the one hand, propose computational data driven steering strategies to overcome input data uncertainty and, on the other hand, to reduce the execution time of the whole prediction process in order to be reliable during real-time crisis. In particular, this work is focused on the description of a Dynamic Data Driven Genetic Algorithm used as steering strategy to automatic adjust certain input data values of forest fire simulators taking into account the underlying propagation model and the real fire behavior.

An asynchronous and iterative load balancing algorithm for discrete load model

Diffusion algorithms are some of the most popular algorithms for dynamic load balancing in which loads move from heavily loaded processors to lightly loaded neighbor processors. To achieve a global load balance in a parallel computer, the algorithm is iterated until the load difference between any two processors is smaller than a specified value. Therefore, one fundamental property to be studied is algorithm convergence. Several analytical works on the convergence of different diffusion load balancing algorithms have been carried out, but they treat loads as non-negative real quantities. In this paper, we describe the Diffusion Algorithm Searching Unbalanced Domains (DASUD) algorithm, which uses loads as non-negative integer values and, unlike existing algorithms, reaches a local balance situation where the maximum load difference between any two processor in the set of neighbor processors for each processor is one load unit. The convergence property of an asynchronous implementation of DASUD using integer loads is proven theoretically.

Parallel ordinary kriging interpolation incorporating automatic variogram fitting

This work introduces a methodology for reducing the execution time of the kriging interpolation method without losing the quality of the model results, as occurs in simplified moving neighborhood solutions. The proposed solution distributes the computation applying parallel programming using MPI (Message Passing Interface) libraries in a HPC (High Performance Computing) environment. For the solution to be automatic and adaptable to different spatial patterns the variogram was automatically fitted; this preliminary modeling step is usually interactive in this interpolation method. The experimental results show the validity of the implemented solution, as it significantly reduces (in one of the examples the execution time decreases from 2h 38min to only 3min) the final execution time of the entire process. The proposed solution is not exclusive to a particular architecture or operating system and can be applied in various environments and spatial resolutions of the generated raster model as well as at different magnitudes of the data to be interpolated.

Performance comparison of dynamic load-balancing strategies for distributed computing

The DASUD (Diffusion Algorithm Searching Unbalanced Domains) algorithm belongs to the nearest-neighbours class and operates in a diffusion scheme where a processor balances its load with all its neighbours. DASUD detects unbalanced domains and performs local exchange of load between processors to achieve global balancing. The DASUD algorithm has been evaluated by comparison with another well-known strategy, namely, the SID (Sender Initiated Diffusion) algorithm across a range of network topologies including ring, torus and hypercube where the number of processors varies from 8 to 128. From the experiments we have observed that DASUD outperforms the other strategy as it provides the best trade-of-between the balance degree obtained at the final state and the number of iterations required to reach such a state. DASUD is able to coerce any initial load distribution into a highly balanced global state and also exhibits good scalability properties.

Injecting Dynamic Real-Time Data into a DDDAS for Forest Fire Behavior Prediction

This work presents a novel idea for forest fire prediction, based on Dynamic Data Driven Application Systems. We developed a system capable of assimilating data at execution time, and conduct simulation according to those measurements. We used a conventional simulator, and created a methodology capable of removing parameter uncertainty. To test this methodology, several experiments were performed based on southern California fires.

A data-driven model for large wildfire behaviour prediction in Europe

Abstract The European Forest Fire Information System (EFFIS) has been established by the Joint Research Centre (JRC) and the Directorate General for Environment (DG ENV) of the European Commission (EC) in close collaboration with the Member States and neighbour countries. EFFIS is intended as complementary system to national and regional systems in the countries, providing harmonised information required for international collaboration on forest fire prevention and fighting and in cases of trans-boundary fire events. However, one missing component in the system is a wildfire behaviour model able to cover the whole Europe. We propose a new general conceptualisation for wildfire prediction. It relies on an array-based and semantically enhanced (Semantic Array Programming) application of the Dynamic Data Driven Application Systems (DDDAS) concept, so as to predict spread of large fires at European level. The proposed mathematical framework is designed to simulate with an ensemble strategy the wildfire dynamics under given sequences of actions for controlling the fire spread and updated data- driven information. First results on data and software uncertainties associated with the problem have been presented with a real case study in Spain.

S 2 F 2 M : statistical system for forest fire management

One of the most serious problems in wildland fire simulators is the lack of precision for input parameters (moisture content, wind speed, wind direction, etc.). In this paper, a statistical method based on a factorial experiment is presented. This method evaluates a high number of parameter combinations instead of considering a single value for each parameter, in order to obtain a prediction which is closer to reality. The proposed methodology has been implemented in a parallel scheme and tested in a Linux cluster using MPI.

Knowledge-guided Genetic Algorithm for input parameter optimisation in environmental modelling

Abstract The need for input parameter optimisation in environmental modelling is long known. Real-time constraints of disaster propagation predictions require fast and efficient calibration methods to deliver reliable predictions in time to avoid tragedy. Lately, evolutionary optimisation methods have become popular to solve the input parameter problem of environmental models. Applying a knowledge-guided Genetic Algorithm (GA) we demonstrate how to speed up parameter optimsation and consequently the propagation prediction of environmental disasters. Knowledge, obtained from historical and synthetical disasters, is stored in a knowledge base and provided to the GA in terms of a knowledge chromosome. Despite of increased loads of knowledge, its retrieval times can be kept near-constant. During GA mutation, ranges of selected parameters are limited forcing the GA to explore promising solution areas. Experiments in forest fire spread prediction show how time-consuming fitness evaluations of the GA could be lowered remarkably to cope with real-time capabilities maintaining the error magnitude.

The Convergence of Realistic Distributed Load-Balancing Algorithms

We give a general model of partially asynchronous, distributed load-balancing algorithms for the discrete load model in parallel computers, where the processor loads are treated as non-negative integers. We prove that all load-balancing algorithms in this model are finite. This means that all load-balancing algorithms based on this model are guaranteed to reach a stable situation at a certain time (which depends on the particular algorithm) at which no load will be sent from one processor to another. With an additional assumption, we prove that the largest load difference between any two processors, in the final stable situation of the load-balancing algorithms in this model, is upper-bounded by the diameter of the topology.