Desarrollo de un modelo de desempeño computacional y planeamiento de capacidad para una aplicación distribuida de búsqueda de variaciones genéticas

Abstract

Nowadays, cloud computing is a model for provision business and technology services, which allows the user standardized services that better suits the needs of their businesses, in addition to reducing costs by charging only for the computational resources used by the application. The virtual infrastructure in cloud computing environments for implementing distributed applications, such as BLAST (Basic Local Alignment Search Tool), must be planned in order to optimize computational performance in terms of time and cost. In recent years, distributed applications have been able to process large amounts of data. The information contained within these data is of great importance; one of the tools that allows extracting this information is MapReduce, a programming model for data analysis created by Google. The main goal of this project is to develop the computational performance model for a distributed application of genetic variations search in different environments. The computational performance model will be able to predict the execution time of a MapReduce BLAST application in a cloud computing environment. This prediction is used to help cloud computing environment users to plan the number of virtual machines, their characteristics, and the time they will use them. To get the computational performance model, the parameters are set to be controlled such as execution time, workload, and the number of virtual machines. After measuring the effects in the execution time caused by the changes in the input parameters, the objective of finding a computational performance model is achieved, which is applied for the prediction of execution times of the BLAST application, and the obtained results are satisfactory. Experiments were made using Amazon Web Services (AWS). The experiments showed a clear relation with the theoretical model, discovering that the model is capable of predicting the execution time of the distributed application of genetic variations search.