Dataflow Oriented Similarity Matching for Scientific Workflows

Abstract

Duplicate and redundant workflows can be avoided by encouraging workflow reuse. In this paper, we present how workflow similarity matching approach can be used to further enhance existing workflow modeling tools. Most existing workflow similarity algorithms cater for controlflow oriented types of workflow which are typically associated with business workflows. The increase presence of scientific workflows that are mainly dataflow oriented calls for workflow similarity matching that caters for these types of workflows instead. We demonstrate here how our work of applying a behavioral analysis technique (taking into consideration the causal footprint of the workflow) that has been used for finding similarity in business workflows perform when use for scientific workflows. The distinction of our technique is the use of data provenance within the scientific workflow model where positional information of the workflow activities are taken in consideration in order to find matching workflow models. Preliminary experiments have shown that our proposed solution provides a viable alternative for matching scientific workflows within multiple scenarios. Furthermore, our suggested approach performs better, particularly with the removal and extension types of modification to the original workflow.

Combining Artificial Intelligence, Ontology, and Frequency-based Approaches to Recommend Activities in Scientific Workflows

“…The work of [8,9,10,11,12,13,14] and [15,16] consider the order of activities, input, output and data provenance. Their limitations are the need of provenance data, since not all Scientific Workflow Management System (SWMS) stores this information.…”

Predição de Falhas em Workflows Científicos em Nuvens baseada em Aprendizado de Máquina

“…A maioria dos trabalhos relacionados ao uso de técnicas de Aprendizado de Máquina para apoiar os usuários de SGWfCs está na composição dos workflows. As tarefas focam em recomendar, na maior parte das vezes, qual sequência de atividades escolher [Yeo and Abidi 2013]. Outros poucos trabalhos focam em utilizar algoritmos de aprendizado de máquina para predizer falhas em execuções.…”