Search

The increased number of extreme wildfires due to climate change generates new challenges in wildfire modelling and prevention actions. One important aspect is to provide current forest fire spread simulation tools with the capability of being faster and more accurate preserving their forecast abilities. In this paper, we propose an alternative forest fire front reconstruction algorithm that could easily be incorporated in the well-known FARSITE fire spread simulator without affecting its functionality. The fire front reconstruction core part in FARSITE represents up to 60% of its total execution time. The proposal alternative algorithm is based on the α-shape algorithm but including some adaptation considering the particular case of forest fire. Meanwhile, the current FARSITE fire front reconstruction algorithm has a complexity of Oleft(n^4 ight), the adapted α-shape algorithm has a complexity of Oleft(nlog{left(n ight)} ight). This new approach has the benefit of being parallelizable using a divide and conquer strategy what will allow to significalty reduce the FARSITE™ execution time without modifying its forecat results.

Computational fire spread models are a useful tool to mitigate the impact of forest fires. Simulators implementing forest fire spread models require several input parameters to deliver their predictions. Wind speed and direction are sensitive parameters that are usually obtained at low resolution (some kilometers). It is necessary to couple a wind field model, that generates a wind field at high resolution (for example, 30 meters). Even in this case of coupled wind field model-forest fire spread simulator, several parameters include a high degree of uncertainty and usually must be calibrated. One of the strategies consists to introduce a Genetic Algorithm (GA) to calibrate the input parameters according to the actual evolution of the fire. GAs are based on running large amount of simulations, and evaluating and comparing the results provided by each simulation. This fact emphasizes how critical are the metrics used to assess the error of the computational forecasts. The goal of this work is to test eight functions to assess the simulation errors in the case of coupled wind field-forest fire spread prediction to study their drawbacks and advantages to determine which is the most relevant error function.

The exploitation of throughput in a parallel application that processes an input data stream is a difficult challenge. For typical coarse-grain applications, where the computation time of tasks is greater than their communication time, the maximum achievable throughput is determined by the maximum task computation time. Thus, the improvement in throughput above this maximum would eventually require the modification of the source code of the tasks. In this work, we address the improvement of throughput by proposing two task replication methodologies that have the target throughput to be achieved as an input parameter. They proceed by generating a new task graph structure that permits the target throughput to be achieved. The first replication mechanism, named DPRM (Data Parallel Replication Mechanism), exploits the inner task data parallelism. The second mechanism, named TCRM (Task Copy Replication Mechanism), creates new execution paths inside the application task graph structure that allows more than one instance of data to be processed concurrently. We evaluate the effectiveness of these mechanisms with three real applications executed in a cluster system: the MPEG2 video compressor, the IVUS (Intra-Vascular Ultra-Sound) medical image application and the BASIZ (Bright and SAtured Images Zone) video processing application. In all these cases, the obtained throughput was greater after applying the proposed replication mechanism than what the application could provide with the original implementation.

The use of high performance computing has been gaining more and more followers in the different branches of Science and Engineering given the potential offered to deal with complex and big problems. However, the parallel programming paradigm involves additional aspects to the merely functional, which could provoke different kinds of bottlenecks in the performance of the applications. Such diff culties may represent critical obstacles, especially for the non-expert users. In this work, we consider the limitations of the automatic development of parallel applications, and the obstacles in the performance tuning process, and we propose a built-in environment that integrates these both processes in a transparent manner. We call our proposal as Environment for automatic development and tuning of Parallel Applications (EPA). EPA follows a cooperative approach: (i) It provides an interface to guide the user in the depiction of the pair problem-solution by completing a form; (ii) such information is automatically formalized in a specif cation; (iii) the specif ed information is automatically used by EPA to instantiate an instrumented skeleton; (iv) the skeleton of the parallel application is a priori enriched in EPA with tuning instrumentation, for monitoring, analysis and tuning actions; (v) the tunable parallel application is automatically generated by EPA; (vi) the user is able to execute the application with no more effort than having completed a form. Clearly, EPA makes transparent the process of code generation and instrumentation, which represents an important advantage especially for non-expert users. Given that the parallelization of any pair problem-solution is hard to generalize, the environment proposes to tackle different parallel problem solvers. The tuning knowledge injected in the generated applications is based on performance models, which makes the decision-making process concise. The Environment could include as problem solvers as needed. In this article, we introduce the problem solver constituted by the Parallel Genetic Algorithms, and — with respect to the tuning process — we consider the improvement of the amount of computational resourses involved in the execution.

Wind field is a critical issue in forest fire propagation prediction. However, wind field calculation is a complex problem that for large terrains involves solving huge linear systems. Solving such systems takes too much time and makes the approach unfeasible in real time operation. To overcome this problem the Schwarz alternating domain decomposition can be applied. Using this method the linear system is decomposed in a set of overlapped subdomains that can be solved in parallel using a Master/Worker paradigm and the wind field calculation time can be significantly reduced.

The classical way of tuning parallel/distributed applications is based on the analysis of the monitoring information obtained from an execution of the application. However, this "measure and modify" approach is not feasible when the applications have a dynamic behavior. In this case, another approach is required to accomplish performance expectations. This paper presents a solution based on the dynamic tuning approach that addresses these issues. In this approach, an application is monitored, its performance bottlenecks are detected and the application is modified automatically during the execution, without stopping, recompiling or re-running it. The introduced modifications adapt the behavior of the application to dynamic variations.

Wind speed and direction are parameters that affect forest fire propagation dramatically. So, an accurate estimation of such parameters is crucial to predict the fire propagation precisely. WindNInja is a wind field simulator that can easily be coupled to a forest fire propagation simulator such as FARSITE. However, wind field simulators present to main drawbacks: They take too much time to compute the wind field and they require a lot of memory. So, a map partitioning strategy has been developed to compute partial wind field maps that can be aggregated afterwards. Each map part can be computed in parallel and the amount of memory required is available in a single node. In this work a methodology to determine the most adequate map partitioning is presented. The map part shape, map part size, amount of overlapping and number of parts have been studied considering execution time and effects on wind field estimation. The results are based on a wide experimentation and are validated with real case scenarios.

There are several parallel applications that are implemented using a Master/Worker parallel/distributed programming paradigm. Applications using this predefined programming structure can be easily implemented using message passing programming libraries (MPI). Moreover, the multicore features present nowadays on CPU architecture can be exploited at the node level by applying thread parallelism (OpenMP). In order to exploit the benefits of both two levels of parallelism, the Master/Worker applications can be implemented as hybrid applications. However, reaching the expected performance indexes is not trivial because there are several parameters (number of nodes, number of threads per node, thread affinity and data distribution among all nodes) that must be tuned for each particular application or even during its execution to reach a successful performance. On the other hand, the application workload may change drastically during successive executions, so those parameters need to be modified according to this behavior. Additionally, cache memory architecture in multicore systems directly influences the performance and this behaviour must be deeply analysed. In this paper we present a proposal to model the performance of hybrid Master/Worker applications on multicore systems considering the issues outlined above. In particular, this model determines dynamically at runtime the configuration of the appropriate number of workers and threads of the hybrid application to achieve the best possible performance.

Defining performance models associated with the application structure has been proven a useful strategy for implementing dynamic tuning tools. However, for extending this strategy to more complex applications (those composed by different structures) it must integrate a policy for the distribution of the resources among the different application components. Consequently, we propose to take advantage of the knowledge of these models and combine them with a resource management policy for obtaining a global model. In this sense, this work constitutes the ongoing effort in the development of performance models for dynamic tuning.

In our research work, we use two Dynamic Data Driven Application System (DDDAS) methodologies to predict wildfire propagation. Our goal is to build a system that dynamically adapts to constant changes in environmental conditions when a hazard occurs and under strict real-time deadlines. For this purpose, we are on the way of building a parallel wildfire prediction method, which is able to assimilate real-time data to be injected in the prediction process at execution time. In this paper, we propose a strategy for data injection in distributed environments.

The increase in the use of parallel distributed architectures in order to solve large-scale scientific problems has generated the need for performance prediction for both deterministic applications and non-deterministic application. In particular, the performance prediction of data-dependent programs is an extremely challenging problem because for a specific issue the input datasets may cause different execution times. Generally, a parallel application is characterized as a collection of tasks and their interrelations. If the application is time-critical it is not enough to work with only one value per task, and consequently knowledge of the distribution of task execution times is crucial. The development of a new prediction methodology to estimate the performance of data-dependent parallel applications is the primary target of this study. This approach makes it possible to evaluate the parallel performance of an application without the need of implementation. A real data-dependent arterial structure detection application model is used to apply the methodology proposed. The predicted times obtained using the new methodology for genuine datasets are compared with predicted times that arise from using only one execution value per task. Finally, the experimental study shows that the new methodology generates more precise predictions.

Parallel computing is a promising approach that provides more powerful computing capabilities for many scientific research fields to solve new problems. However, to take advantage of such capabilities it is necessary to ensure that the applications are successfully designed and that their performance is satisfactory. This implies that the task of the application designer does not finish when the application is free of functional bugs, and that it is necessary to carry out some performance analysis and application tuning to reach the expected performance. This application tuning requires a performance analysis, including the detection of performance bottlenecks, the identification of their causes and the modification of the application to improve behavior. These tasks require a high degree of expertise and are usually time consuming. Therefore, tools that automate some of these tasks are useful, especially for non-expert users. In this paper, we present three tools that cover different approaches to automatic performance analysis and tuning. In the first approach, we apply static automatic performance analysis. The second is based on run-time automatic analysis. The last approach sets out dynamic automatic performance tuning.

Performance is the reason for parallel computing. Despite years of work, many applications achieve only a few percentage of theoretical peak. Performance measurement and analysis tools exist to identify the problems with current programs and systems. Performance Prediction is intended to identify issues in new code or systems before they are fully available. These two topics are closely related since most prediction requires data to be gathered from measured runs of program (to identify application signatures or to understand the performance characteristics of current machines).

A static scheduling strategy based on task replication was previously proposed by the authors (1990). This method allowed the possibility that certain nodes of the directed acyclic graph (DAG) could vary their execution times within a range of known values. The present work extends that scheduling strategy using a new representation model of a parallel program, the directed graph model. This model allows the inclusion of loops and conditionals explicitly in the topology of the graph. The extension of the model takes advantage of the explicit parallelism existing within loops and conditional structures that is hidden in the DAG model. The task replication strategy is adapted to the new situation to obtain a static assignment that reproduces the dynamic behavior of the graphs without task migration during application run time. >

Actual behaviour of parallel programs is of capital importance for the development of an application. Programs will be considered matured applications when their performance is over acceptable limits. Traditional parallel programming forces the programmer to understand the enormous amount of performance information obtained from the execution of a program. In this paper, we propose an automatic analysis tool that lets the programmers of applications avoid this difficult task. This automatic performance analysis tool main objective is to find poor designed structures in the application. It considers the trace file obtained from the execution of the application in order to locate the most important behaviour problems of the application. Then, the tool relates them with the corresponding application code and scans the code looking for any design decision which could be changed to improve the behaviour.

Data assimilation in weather forecasting is a well-known technique used to obtain an improved estimation of the current state of the atmosphere (analysis). The Meteorological Service of Catalunya (SMC) is seeking for a real time high resolution analysis of surface parameters over Catalonia (north-east of Spain), in order to know the current weather conditions at any point of that region. For this purpose, a comparative study among several data assimilation experiments based on LAPS (Local Analysis and Prediction System) and STMAS (Space-Time Multiscale Analysis System) and multi-regression technique designed at SMC, has been performed to determine which one delivers best results. The comparison has been done using as true state independent observational data provided by the Spanish Meteorological State Agency (Agencia Estatal de METeorologia, AEMET). The results show that the multi-regression technique provides more accurate analyses of temperature and relative humidity than the LAPS/STMAS experiments, mainly due to the fact that multi-regression methodology only uses observations and consequently the model biases are avoided.