On Wednesday, May 4th, 2016, GMAP researcher Carlos Maron presented his Master's Degree Research Plan at the Faculty of Informatics of the Pontifical Catholic University of Rio Grande do Sul.
The proposal, called "BenSP (Benchmark for Stream Parallelism): A Benchmark for Stream Parallelism of Multi-Core Architectures", was presented at 10:30 am in room 513 of building 32, with an examining board consisting of Professors Luiz Gustavo Fernandes and Tiago Ferreto.
Summary: S tream applications are found in various computer systems of our daily lives. For example, video and image processing, monitoring systems, social networking, among others. The stream is characterized as a set of streaming elements (eg, picture, number, structure, etc.), and is in its most part infinite. However, it may be finite when there is a stopping criterion. The parallelization of these applications becomes a challenge in order to find the best performance configuration. Each program has a particular feature for dealing with this continuous flow, because several properties are variable. For example, the stage may have diverse computations (with or without internal state), the stream elements may be of different sizes, among other properties. A configurable benchmark for this type of application becomes an attraction to find the best parallelism strategy in a given architecture. Essentially, because traditional programming interfaces such as BBT and FastFlow that allow parallelization stream, need a semantic processing of the sequential code; since each strategy to be tested is necessary to reschedule and restructure source code, a benchmark becomes a common feature to collect metrics, since the results are achieved more easily. Generally, a benchmark resembles a real application, so, during testing, the actual scenario aspects can be anticipated. Bearing in mind the complexity of parallelization of a stream and the need for a benchmark for this scenario, we suggest in this work the BenSP (Benchmark for Stream Parallelism).
This benchmark seeks to contemplate important attributes of the stream parallelism in an adjustable manner in a multi-core architecture, for example, Input, Output and Replicate. Then, BenSP will serve as a benchmark for simulating stream parallelized applications in these architectures,avoiding the programming effort.
GMAP researcher Carlos Maron presenting his Master's Degree Research Plan.