Alvaro Luiz Fazenda

A Comparative Analysis of Load Balancing Algorithms Applied to a Weather Forecast Model

Among the many reasons for load imbalance in weather forecasting models, the dynamic imbalance caused by localized variations on the state of the atmosphere is the hardest one to handle. As an example, active thunderstorms may substantially increase load at a certain time step with respect to previous time steps in an unpredictable manner – after all, tracking storms is one of the reasons for running a weather forecasting model. In this paper, we present a comparative analysis of different load balancing algorithms to deal with this kind of load imbalance. We analyze the impact of these strategies on computation and communication and the effects caused by the frequency at which the load balancer is invoked on execution time. This is done without any code modification, employing the concept of processor virtualization, which basically means that the domain is over-decomposed and the unit of rebalance is a sub-domain. With this approach, we were able to reduce the execution time of a full, real-world weather model.

Acc-Motif: accelerated network motif detection

Network motif algorithms have been a topic of research mainly after the 2002-seminal paper from Milo et al. [1], which provided motifs as a way to uncover the basic building blocks of most networks. Motifs have been mainly applied in Bioinformatics, regarding gene regulation networks. Motif detection is based on induced subgraph counting. This paper proposes an algorithm to count subgraphs of size k + 2 based on the set of induced subgraphs of size k. The general technique was applied to detect 3, 4 and 5-sized motifs in directed graphs. Such algorithms have time complexity O(a(G)m), O(m2) and O(nm2), respectively, where a(G) is the arboricity of G(V, E). The computational experiments in public data sets show that the proposed technique was one order of magnitude faster than Kavosh and FANMOD. When compared to NetMODE, acc-Motif had a slightly improved performance.

Accelerated Motif Detection Using Combinatorial Techniques

Network motif algorithms have been a topic of research mainly after the 2002-seminal paper from Milo et al, that provided motifs as a way to uncover the basic building blocks of most networks. This article proposes new algorithms to exactly count isomorphic pattern motifs of size 3 and 4 in directed graphs. The algorithms are accelerated by combinatorial techniques. Let G(V, E) be a directed graph with m=|E|. We describe an O(m√m) time complexity algorithm to count isomorphic patterns of size 3. To counting isomorphic patterns of size 4, we propose an O(m2) algorithm. The new algorithms were implemented and compared with Fanmod motif detection tool. The experiments show that our algorithms are expressively faster than Fanmod. We also let our tool to detect motifs, the ACC-MOTIF, available in the Internet.

Lyapunov exponents and adaptive mesh refinement for high-speed flows using a discontinuous Galerkin scheme

This paper proposes two important improvements to shock-capturing strategies using a discontinuous Galerkin scheme, namely, accurate shock identification via finite-time Lyapunov exponent (FTLE) operators and efficient shock treatment through a point-implicit discretization of a PDE-based artificial viscosity technique. The advocated approach is based on the FTLE operator, originally developed in the context of dynamical systems theory to identify certain types of coherent structures in a flow. We propose the application of FTLEs in the detection of shock waves and demonstrate the operators ability to identify strong and weak shocks equally well. The detection algorithm is coupled with a mesh refinement procedure and applied to transonic and supersonic flows. While the proposed strategy can be used potentially with any numerical method, a high-order discontinuous Galerkin solver is used in this study. In this context, two artificial viscosity approaches are employed to regularize the solution near shocks: an element-wise constant viscosity technique and a PDE-based smooth viscosity model. As the latter approach is more sophisticated and preferable for complex problems, a point-implicit discretization in time is proposed to reduce the extra stiffness introduced by the PDE-based technique, making it more competitive in terms of computational cost.

Improving the Scalability of an Operational Scientific Application in a Large Multi-core Cluster

Currently, High-Performance Computers use nodes with a tendency of an increasing number of cores per chip. In this scenario, enhancing scalability of an existing application requires a comprehensive approach, since system parameters such as memory per core and I/O speeds increase slower with time than cores per chip. This work describes the enhancements incorporated in BRAMS - a regional weather forecasting model - to reach a target execution time using 9,600 cores. We show that some common coding techniques may prevent scalability and that I/O and memory are constraints as core counts increase.

Evaluating Volunteers' Contributions in a Citizen Science Project

Today, with freely available data coming from different satellites and instruments, efficient algorithms for image classification, and increased connectivity and computing power, together with international policy initiatives, such as the United Nations Framework Convention of Climate Change Reducing Emissions from Deforestation and Forest Degradation (REDD) program, more and more countries are starting to invest in their own national forest monitoring schemes. Still, tropical forests remain under threat worldwide. Recently a citizen science project that enables citizens around the globe to be involved in forest monitoring tasks has been proposed. Called Forest Watchers (www.forestwatchers.net), its main goal is to allow volunteers (many of them with no scientific training) around the globe, with their own smartphones, tablets and notebooks, to review satellite images of forested regions, and confirm whether automatic assignments of forested and deforested regions are correct. Inspected images are then sent to a central database where results are integrated to generate up-to-date deforestation maps. This approach offers a low-cost way to both strengthen the scientific infrastructure and engage members of the public in science. Here we describe the procedures developed within the scope of the Forest Watchers project to assess the tasks performed by the volunteers. These procedures have been evaluated with data of one of the projects preliminary applications. Called BestTile, it asks volunteers to select among several images of the same area, which one has the least cloud cover. Results from more than 500 volunteers show that with simple statistical tests it is possible to attain a triple goal: to increase the overall efficiency of the data collecting tasks, by reducing the required number of volunteers per task, to identify malicious behavior and outliers, and to motivate volunteers, to continue their contributions.

Parallel Algorithm for Dynamic Community Detection

Many real systems can be naturally modeled by complex networks. A complex network represents an abstraction of the system regarding its components and their respective interactions. Thus, by scrutinizing the network, interesting properties of the system can be revealed. Among them, the presence of communities, which consists of groups of densely connected nodes, is a significant one. For instance, a community might reveal patterns, such as the functional units of the system, or even groups correlated people in social networks. Albeit important, the community detection process is not a simple computational task, in special when the network is dynamic. Thus, several researchers have addressed this problem providing distinct methods, especially to deal with static networks. Recently, a new algorithm was introduced to solve this problem. The approach consists of modeling the network as a set of particles inspired by a N-body problem. Besides delivering similar results to state-of-the-art community detection algorithm, the proposed model is dynamic in nature; thus, it can be straightforwardly applied to time-varying complex networks. However, the Particle Model still has a major drawback. Its computational cost is quadratic per cycle, which restricts its application to mid-scale networks. To overcome this limitation, here, we present a novel parallel algorithm using many-core high-performance resources. Through the implementation of a new data structure, named distance matrix, was allowed a massive parallelization of the particles interactions. Simulation results show that our parallel approach, running both traditional CPUs and hardware accelerators based on multicore CPUs and GPUs, can speed up the method permitting its application to large-scale networks.

Numerical weather model BRAMS evaluation on many-core architectures: a micro and macro vision

This paper investigates the performance of a weather forecasting application Brazilian developments on the regional atmospheric modelling system - BRAMS on high performance computing HPC clusters with a multi-core architecture. We simulated atmosphere conditions over South America for 24 hours ahead using the BRAMS, aiming to understand the impact of different architectural configurations on performance and scalability. Our analyses consider execution in intra-node and inter-node configurations of a cluster with 24 cores per node. Results reveal differences in the BRAMS performance caused by interconnection. The BRAMS may get better performance by using a newer version of MPI library implementation one-copy schema and improving spatial resolution.

Characterizing Communication Patterns of Parallel Programs Through Graph Visualization and Analysis

Characterization of communication patterns of parallel programs has been used to better understand the behavior of such programs as well as to predict performance of large scale applications. This characterization could be performed by observing some communication attributes like volume or spatial characteristics of message passing parallel applications in different scenarios. This paper describes a methodology to characterize parallel communication patterns using a graph visualization tool in addition to a traditional monitoring tool that generates trace files. Graph visualization tools are commonly used to analyze large network connections existent in a variety of social or natural structures. Although, since it is possible to represent large scale parallel programs as graphs of communicating processes, this paper proposes a methodology that takes advantage of such kind of tool to aid in characterize communication patterns.

XV Brazilian Symposium on High Performance Computational Systems (WSCAD 2014)

We are very pleased to welcome you to this edition of the Journal of Physics: Conference Series. This is a special issue for WSCAD 2014 (XV Brazilian Symposium on High Performance Computational Systems) which comprises a set of seven papers very carefully selected from the best papers from the WSCAD conference in 2014. The authors of the selected papers submitted an extended version of their work and then all the papers went through a new review process. We are thankful to the authors for their contributions to this special issue.