Authors:
Paulo Costa
1
;
Edward David Moreno Ordonez
2
and
Jean Teixeira de Araujo
3
Affiliations:
1
Coordenadoria de Análise e Desenvolvimento de Sistemas (CADS), Instituto Federal de Sergipe, Aracaju, Brazil
;
2
Departamento de Computação, Universidade Federal de Sergipe, São Cristovão, Brazil
;
3
Departamento de Informática, Universidade Federal do Agreste de Pernambuco, Garanhuns, Brazil
Keyword(s):
Software Aging, Software Rejuvenation, SAR, Predict Model, Cloud Computing, Edge Computing, Fog Computing, Systematic Mapping Literature, SML.
Abstract:
This article presents a Systematic Literature Mapping (SLM), related to software aging and rejuvenation prediction models. The study highlights the importance of these models, due to the high cost of software or service downtime in IT datacenter environments. To mitigate this impact and seek greater reliability and availability of applications and services, software aging prediction and proactive rejuvenation are significant research topics in the area of Software Aging and Rejuvenation (SAR). Costs are potentially higher when rejuvenation actions are not scheduled. Various prediction models have been proposed for over twenty-five years, with the aim of helping to find the ideal moment for rejuvenation, in order to optimize the availability of services, reduce downtime and, consequently, the cost. However, the scope of this study was limited to a survey of the last fifteen years of models with a measurement-based prediction strategy. These models involve monitoring and collecting dat
a on resource consumption over time, from a running computer system. The collected data is used to adjust and validate the model, allowing the prediction of the precise moment of the aging phenomenon and the consequent rejuvenation action of the software. In addition to providing a baseline from the compiled prediction models, identifying gaps that could encourage future research, particularly in the areas of machine learning or deep learning, the research also contributed to clarifying that hybrid algorithms based on Long Short-Term Memory (LSTM) are currently situated at the highest level of prediction models for software aging, with recent highlights for two variants: the Gated Recurrent Unit (GRU) and the Bidirectional Long Short Term Memory (BiLSTM). Objectively, in response to the research questions, the article also contributes by presenting, through tables and graphs, trends and consensus among researchers regarding the evolution of prediction models.
(More)