He-lasic : um modelo de elasticidade em nuvem baseado em duas camadas para acelerar o cálculo de best-fit em sistemas de substituição filogenética

Abstract

Currently there are few examples of areas of knowledge that don’t benefit from technological resources to increase productivity. Also being one of the beneficiaries of this trend, bioinformatics, which emerges from the combination of knowledge in biology, computer science and medicine, has established techniques, algorithms and databases that have become fundamental for the execution of research, whether in the academic or professional spheres. Recent advances such as Next Generation Gene Sequencing (NGS) technologies have brought abundance of data to bioinformatics and phylogenetics, which focuses on the study of genealogical relationships between organisms, increasing the quality of studies once limited to manual observations only. With the objective of increasing knowledge about bioinformatics and phylogenetics, in collaboration with the Biology laboratory of UNISINOS, we sought out to carry a mapping of the state of the art through a literature review that contributed to the formulation of a taxonomy capable of classifying the software projects found in terms of their categories, methods and purposes. We conclude, on the basis of the bibliography, that in contrast to the recent advances in phylogenetics, software projects are technologically outdated, requiring investments in modernization mainly regarding the application of more recent techniques such as resource elasticity, load balancing and GPU acceleration. Through the He–lastic model an optimization approach is propos ed for the jModelTest project which encompasses the use of resource elasticity in a cloud computing environment, resulting in a two layer elasticity architecture, base on FaaS and Container Orchestration technologies. The proposed model was submitted to a series of evaluation scenarios that allowed for comparison with the current state of jModelTest and resulted in an overhead, encompassing execution time and financial cost, between 6% to 16%, depending on the number of resources allocated. Among the contributions achieved by our proposal, the He–lastic model is highlighted, with its dual elasticity layers approach, which avoids idle costs, as well as a cost study detailing the interaction between elasticity layers, and a taxonomy which was fundamental in the classification of the papers included in the state of the art survey. In the future, we envision efforts to optimize the transition between elasticity layers, which presented significant efficiency losses during the evaluation, as well as improvements to the prototype and our model in order to extract greater efficiency gains.