Captura de proveniência assíncrona em simulações computacionais

Abstract

Large-scale computational simulations are computational experiments increasingly more processing intensive. Users and developers of this type of simulation generally analyze data during simulation execution. This is not a trivial task since largescale simulations are often performed in high-performance processing environments and can produce a large volume of data. Existing solutions, as DfAnalyzer, use provenance data to assist analysis with success. However, these systems use synchronous approaches to gather data that makes difficult to set up it and, mainly, interferes in the performance of the computational simulation. This dissertation proposes an approach to asynchronously collect provenance data making it available for analysis during the execution of the simulation with the least possible delay. In order to evaluate the proposed strategies, a tool, Asynchronous Dataflow Analyzer. This implementation extends DfAnalyzer to use the proposed asynchronous approach and to simplify the configuration process by making the prospective provenance definition process more flexible. The experimental results, with a soils sedimentation simulation, show that the tool is able to meet the needs of users of large-scale computational simulations with lower overloads than similar tools.