Research on Computing Tasks

Jan Bielecki

supervisor: Michał Śmiałek

The research focuses on the Big Data field's part, which is computing tasks (programs written in the functional paradigm and executions of such programs).

I conduct my research on the BalticLSC system but keep it the most general and one can use the results in different systems or cases. BalticLSC is a platform that uses a new visual language (CAL) to solve some of the High-Performance Computing problems. It uses high-level abstractions to define the flow of data and the execution of computation modules in a distributed computation environment.

The first part of my research focuses on the estimation of execution time for computing tasks. The proposed approach is to create machine learning models using historical data. The models use program metadata and parameters of the run-time environment as their explanatory variables. Moreover, while creating the model, one can expand the set of variables with additional parameters of the specific programs. Each module (computing task) has its model.

I researched the process of training and validation for several different computation modules and discussed the suitability of the proposed models for ACET estimation in various computing environments. Experiments were carried out using SVR and KNN algorithms.

Further work will concentrate on static analysis of programs code. Based on the control flow graphs of the programs, one can translate its structure to vector. Such vectors could be an issue as the parts of input data for the already checked models or help build more sophisticated models.