Khái niệm cốt lõi
Analyzing the scalability and parallelization of local search algorithms for the Satisfiability problem using runtime distributions.
Tóm tắt
This paper delves into predicting parallel performance by analyzing sequential runtime distributions. It introduces a model based on order statistics to estimate parallel execution. The study focuses on two SAT solvers, Sparrow and CCASAT, comparing predicted and empirical performances up to 384 cores. Results show that the model accurately predicts performance close to actual data. Different types of instances exhibit varying behaviors approximated by exponential or lognormal distributions. The analysis extends to crafted instances, where Sparrow shows linear speedup with nearly optimal scaling as core count increases.
Thống kê
"We apply this approach to study the parallel performance of two SAT local search solvers, namely Sparrow and CCASAT, and compare the predicted performances to the results of an actual experimentation on parallel hardware up to 384 cores."
"Moreover, extensive experimental results (up to 384 cores) using state-of-the-art local search solvers showed that the predicted execution times and speedups accurately match the empirical data and performance."
"The main contributions of this paper are as follows."
"All the experiments were performed on the Grid’5000 platform, the French national grid for research."
"In order to obtain the empirical data for the theoretical distribution (predicted by our model from the sequential runtime distribution), we performed 500 runs of the sequential algorithm."
Trích dẫn
"We propose a framework to estimate the parallel performance of a given algorithm by analyzing the runtime behavior of its sequential version."
"Results show that the model accurately matches the parallel performance of empirical experiments up to 384 cores."
"The analysis extends to crafted instances, where Sparrow shows linear speedup with nearly optimal scaling as core count increases."