A Systematic Mapping on Software Aging and Rejuvenation Prediction
Models in Edge, Fog and Cloud Architectures
Paulo do Amaral Costa
1 a
, Edward David Moreno Ordonez
2 b
and
Jean Carlos Teixeira de Araujo
3 c
1
Coordenadoria de An
´
alise e Desenvolvimento de Sistemas (CADS), Instituto Federal de Sergipe, Aracaju, Brazil
2
Departamento de Computac¸
˜
ao, Universidade Federal de Sergipe, S
˜
ao Cristov
˜
ao, Brazil
3
Departamento de Inform
´
atica, Universidade Federal do Agreste de Pernambuco, Garanhuns, Brazil
Keywords:
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 pre-
diction 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 data 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 Bidirec-
tional 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.
1 INTRODUCTION
According to recent research (Yue et al., 2020), the
annual cost of system downtime in Information Tech-
nology (IT) environments is approximately US$ 26.5
billion. Reducing this costly impact and maximiz-
ing the reliability and availability of applications and
services justify efforts to predict software aging and
proactive and planned rejuvenation action. These top-
ics are highly relevant and of great interest to re-
searchers in the field of Software Aging and Rejuve-
a
https://orcid.org/0009-0004-4252-2963
b
https://orcid.org/0000-0002-4786-9243
c
https://orcid.org/0000-0002-1688-4782
nation (SAR) (Araujo et al., 2011; Cotroneo et al.,
2014; Di Sanzo et al., 2015; Avresky et al., 2017;
Umesh et al., 2017; Liu et al., 2019; Wang and Liu,
2020; Tan and Liu, 2021; Oliveira et al., 2021).
This article presents a Systematic Literature Map-
ping (SLM) conducted to address research questions
specifically related to software aging and rejuvenation
prediction models in edge, fog and cloud computing
environments.
The SLM collected evidence on work on software
aging prediction systems in four scientific literature
databases, Figure 1, over the last 15 years (period
from 2009 to 2023).
The retrieved documents answered several re-
search questions, however the SLM was guided by
Costa, P., Ordonez, E. and Teixeira de Araujo, J.
A Systematic Mapping on Software Aging and Rejuvenation Prediction Models in Edge, Fog and Cloud Architectures.
DOI: 10.5220/0012633900003690
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 26th International Conference on Enterprise Information Systems (ICEIS 2024) - Volume 1, pages 933-942
ISBN: 978-989-758-692-7; ISSN: 2184-4992
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
933

the following main question: “What predictive soft-
ware aging models have been proposed in Edge, Fog
and Cloud Computing environments?”
The literature search revealed that algorithms
based on decision trees and time series were the most
cited in the selected studies. However, in the last
years, the Long Short-Term Memory (LSTM) convo-
lutional neural network, the present state of the art
in temporal series prediction, demonstrated the most
promising results when applied to software aging pre-
diction.
LSTM is a deep learning algorithm often applied
to sequential data sets, such as time series, and as soft-
ware aging is something that occurs over time, its use
is appropriate. The research contributed to clarify that
the proposed hybrid models based on LSTM (Battisti
et al., 2022; Shi et al., 2023; Jia et al., 2023) are at the
most evolutionary level of prediction models for soft-
ware aging, with more recent highlights for two vari-
ants: the Gated Recurrent Unit (GRU) and the Bidi-
rectional Long Short - Term Memory (BiLSTM).
The GRU has a simpler structure that provides
better performance than the LSTM model and with
slightly greater or equivalent accuracy. BiLSTM or
Bidirectional LSTM, composed of two LSTMs, has
the ability to analyze future and past timestamp in-
puts, with its backward and forward direction lay-
ers, being a more powerful algorithm than the original
unidirectional LSTM and, therefore, the prediction of
this model tends to be better.
Figure 1: Articles selected per source.
The results of this study, based on the compiled
models, also contribute to identifying gaps that can
stimulate future research proposals, especially those
based on deep learning, whether or not considering
the interdependence of aging indicator variables.
Furthermore, the remaining sections of the article
are organized as follows. Section 2 discusses the def-
initions of software aging and rejuvenation, as well
as prediction strategies. Section 3 defines the SLM
method used in this study. Section 4 summarizes the
results obtained from the analysis of selected articles
through tables and graphs. Section 5 addresses threats
to the validity of the study. Finally, Section 6 presents
the conclusions regarding this study.
2 THEORETICAL FOUNDATION
2.1 Software Aging and Rejuvenation
Software aging refers to an increase in system per-
formance degradation that occurs over its operational
lifespan. It is a cumulative process caused by software
errors that gradually deplete system resources, with-
out immediately resulting in a failure (Grottke et al.,
2008; Tang et al., 2020; Oliveira et al., 2021).
The symptoms of software aging are perceived or
evidenced through real-time monitoring of system re-
sources, which are commonly referred to as “aging in-
dicators” (Grottke et al., 2008; Tang et al., 2020). The
main indicators include primary memory, swap vir-
tual memory, CPU utilization, and secondary memory
usage.
Software rejuvenation is a proactive technique
that aims to alleviate the effects of software aging
(Cotroneo et al., 2014). Its goal is to clean up the
internal degradation state of the system, preventing
more severe failures such as crashes or malfunctions,
and restoring its performance (Sudhakar et al., 2014;
Cotroneo et al., 2014; Umesh et al., 2017).
The most common rejuvenation techniques in-
volve restarting a specific software component or re-
booting the entire system (Araujo et al., 2011; Araujo
et al., 2014; Cotroneo et al., 2014; Battisti et al.,
2022). In the latter case, for example, in the case of
swap space, it is not cleared after restarting the aging-
causing application but only after a complete restart of
the operating system (Simeonov and Avresky, 2010).
One of the typical challenges of a rejuvenation ap-
proach is the software downtime, as the service or ap-
plication becomes unavailable during the entire restart
process (Araujo et al., 2011; Wang and Liu, 2020;
Oliveira et al., 2021; Battisti et al., 2022). Therefore,
a software rejuvenation strategy should be carefully
planned to avoid more severe failures and optimize
downtime (Cotroneo et al., 2014; Tan and Liu, 2021).
The costs of downtime are higher when it is not
scheduled (Cotroneo et al., 2014), and predicting the
failure time of aging and planning when to perform
the rejuvenation action are highly relevant in the SAR
field (Cotroneo et al., 2014; Liu et al., 2019; Wang
and Liu, 2020). These efforts aim to improve service
availability and reduce overall downtime.
ICEIS 2024 - 26th International Conference on Enterprise Information Systems
934

2.2 Prediction Strategies
Currently, there are three main categories of proac-
tive rejuvenation prediction strategies aimed at mit-
igating the effects of software aging (Simeonov and
Avresky, 2010; Cotroneo et al., 2014; Liu et al.,
2019). These categories are: a) model-based strategy,
b) measurement-based strategy, and c) hybrid strat-
egy.
However, the scope of the compilation in this
study is limited to the second type of strategy. For the
measurement-based strategy, the first step involves
monitoring and collecting data on system resource
consumption during operation (Di Sanzo et al., 2015).
In a subsequent stage, time series analysis (Araujo
et al., 2011; Araujo et al., 2014; Umesh et al., 2017),
as well as machine learning algorithms (Simeonov
and Avresky, 2010; Sudhakar et al., 2014; Di Sanzo
et al., 2015; Avresky et al., 2017), including their sub-
field of deep learning (Yue et al., 2020; Tan and Liu,
2021), or deep learning in hybrid approach (Liu et al.,
2019; Battisti et al., 2022; Shi et al., 2023; Jia et al.,
2023), are used to process the collected data. The pre-
diction model is then trained to accurately predict the
occurrence of software aging as precisely as possible
(Tan and Liu, 2021).
When it comes to software aging and rejuvenation
for cloud services, prediction methods based on mea-
surement strategies tend to be more promising (Liu
et al., 2019).
3 METHODOLOGY
A Systematic Literature Mapping (SLM), among
other characteristics, aims to provide an overview of
a specific research area, identify utilized techniques,
and facilitate the discovery of gaps that can support
future investigations through its systematic methodol-
ogy (Kitchenham et al., 2011; Petersen et al., 2015).
This article is the result of an SLM focused on pri-
mary studies published on software aging prediction
systems.
3.1 Research Questions
The research adopted the systematic PICOC analysis
(Population, Intervention, Comparison, Results, Con-
text). It is the starting point for developing the re-
search questions and search string. PICOC is a proce-
dure used to describe the five elements related to the
identified problem and to structure the main question
of a research.
Following the protocol recommended (Kitchen-
ham et al., 2011; Petersen et al., 2015), when formu-
lating the PICOC framework, it was obtained:
• Population (Group from which evidence is col-
lected). Software aging and rejuvenation predic-
tion systems;
• Intervention (Action applied in the empirical
study). Synthetic analysis of prediction models;
• Comparison (Parameters with which the interven-
tion is compared). Regression model, classifica-
tion model;
• Outcomes (desired outcomes). Prediction tech-
niques or algorithms, machine learning and deep
learning algorithms, aging indicators and strate-
gies, environments;
• Context (Segment in which the population is lo-
cated). Cloud, Fog and Edge Architectures.
It is clear that research questions must take into
account the following points of view: Population, In-
tervention and Results, the objective of which is to
answer the components: Comparison and Context.
Based on this principle and the research interest de-
fined through PICOC analysis, it became possible
to formulate the main research question (MQ) that
guided the present study: “What predictive software
aging models have been proposed in Edge, Fog and
Cloud Computing environments?”
To answer the research interest and provide a
broader scope, the MQ was broken down and dis-
sected into the following research questions (RQs):
• RQ1 - Which techniques or algorithms for soft-
ware aging prediction were used in the experi-
ments?
• RQ2 - Which aging indicators were analyzed in
the experiments?
• RQ3 - What was the method used to obtain the
aging indicator data in the experiments?
• RQ4 - What aging strategies were adopted in the
experiments?
• RQ5 - Which virtualization environments were
used in data acquisition or in the aging experi-
ments?
• RQ6 - Which network architecture has the highest
incidence of studies on software aging prediction
systems?
3.2 Search String
Once the research questions were formulated, the next
step was to create the search string, which enabled the
A Systematic Mapping on Software Aging and Rejuvenation Prediction Models in Edge, Fog and Cloud Architectures
935

retrieval of evidence in the literature. The terms (syn-
onymous descriptors or key words) identified and re-
lated to each of the components of the PICO strategy
and the use of Boolean operators (AND, OR, NOT)
for each of the four components of the strategy, in-
terrelating the sentences (P) AND (I) AND (C) AND
(O), provided the construction of the search string, as
shown in Figure 2.
("software aging" OR "sofware rejuvenation")
AND
("edge computing" OR "fog computing" OR "cloud computing")
AND
(algorithm OR "machine learning" OR "deep learning" OR "neural network"
OR ANN OR CNN OR RNN OR LSTM OR ConvNET OR "time series")
Figure 2: Query string.
The search string was submitted directly to four
literature databases: ACM Digital Library; IEEE
XPlore; Scopus; Web of Science. A total of seventy
two articles were retrieved, as shown in Figure 3.
After thorough reading, only seventeen articles
were accepted for data extraction. Sixteen articles
were duplicates and thirty-nine articles were rejected.
3.3 Inclusion and Exclusion Criteria
Inclusion and exclusion criteria are used to exclude
studies that are not relevant to answering the research
questions (Petersen et al., 2015). The inclusion crite-
ria (IC) adopted were:
• IC1 - The publication’s objective must be related
to software aging and rejuvenation prediction.
• IC2 - The publication must describe a model,
method, or technique for software aging predic-
tion.
The exclusion criteria (EC) adopted were:
• EC1 - The publication does not have an abstract.
• EC2 - The study is published only as an abstract.
• EC3 - The publication is not written in English.
• EC4 - The publication is an older version of an-
other already considered.
• EC5 - The publication is not a primary study.
• EC6 - The publication does not have full-text
availability.
• EC7 - The publication does not pertain to software
aging and rejuvenation prediction systems.
• EC8 - Access to the publication was not possible.
• EC9 - The document does not address
measurement-based prediction strategy.
• EC10 - The document has not been published in
the last 15 years.
Scopus
(15)
Web of
Science
(11)
ACM Digital
Library
(27)
IEEE
Xplore
(19)
Total Papers
Selected: 72
Rejected: 34
Duplicated: 16
Accepted: 22
Full Reading
Rejected: 5
Accepted: 17
Data Analysis
MQ
(17)
RQ1
(17)
RQ2
(17)
RQ3
(17)
RQ4
(9)
RQ5
(17)
RQ6
(17)
Figure 3: Document selection and data extraction process.
4 RESULTS AND DISCUSSION
4.1 Preliminary Results
In this preamble, additional secondary information
is presented, however it is very important for under-
standing this mapping.
Seventeen articles related to predictive models of
software aging and rejuvenation were selected.
Figure 4 presents a comparison over the years be-
tween the documents initially selected and the doc-
uments accepted after full reading. Both curves are
clearly multimodal and point to the years 2014, 2016,
2020 and 2021 with the highest volumes of selected
publications and the years 2014, 2017, 2020, 2021
and 2023 with the most articles finally accepted.
The graph in Figure 5 crosses the number of arti-
cles selected and articles finally accepted per source.
Highlights include the Web of Science, IEEE Xplorer
and Scopus databases, which together accounted for
94.1% of the articles finally accepted. The ACM Dig-
ital Library source, despite contributing the largest
volume of selected articles (37.5%, Figure 1), had
only 1 article finally accepted.
ICEIS 2024 - 26th International Conference on Enterprise Information Systems
936

Figure 4: Number of articles published per year according
to the search string.
The same publication may involve affiliations of
researchers from different countries. Therefore, Asia,
Europe, South America and North America constitute
the continents of the authors of the 17 accepted pub-
lications.
Figure 5: Number of articles selected and articles accepted
per source.
Figure 6 highlights the researchers from China,
Germany and Brazil, with 8, 4 and 4 participations
in accepted publications on software aging prediction
models.
4.2 Answers to Research Questions
It is worth noting that no publications with experi-
ments specifically related to the Fog and Edge Com-
puting segments were found using the search string.
Only eight articles did not answer research ques-
tion number 4. All other RQs were answered in full
by all accepted documents (see Figure 3).
The answers to all investigated questions will be
presented sequentially through tables, including the
main research question.
Figure 6: Participation of the country of affiliation in publi-
cations.
4.2.1 MQ - What Predictive Software Aging
Models Have Been Proposed in Edge, Fog
and Cloud Computing Environments?
Table 1 shows that around 88% of the works pro-
pose prediction models based on regression algo-
rithms (15 references), in order to estimate Time-to-
Failure (TTF) or Remaining Time-To-Failure (RTTF),
which is a prognosis of the remaining useful life of the
system for performance degradation or sudden down-
time, resulting from resource exhaustion and/or accu-
mulations of numerical errors or aging-related bugs,
caused by continuous execution over a long period.
It is observed that only 2 works refer to classifi-
cation algorithms. Simeonov’s work (Simeonov and
Avresky, 2010) uses a machine learning algorithm
based on a decision tree for a binary “yes/no” clas-
sification. Machine learning was developed from his-
torical data. If the value is “yes”, the corresponding
machine needs to be rejuvenated.
Yue (Yue et al., 2020) proposes a deep learning
method to predict the phenomenon of microservice
aging and a rejuvenation policy. The network archi-
tecture is the CNN+LSTM model, where, in the net-
work input layer, each neuron represents a microser-
vice. The output layer activated by the Softmax func-
tion is a vector of 10 categories, where each class re-
spectively represents the probability of quality of ser-
vice (QoS) violations. When the QoS violation proba-
bility value is high and CPU, memory, and disk usage
exceeds standard limits, the microservice is consid-
ered obsolete and is rejuvenated.
4.2.2 RQ1 - Which Techniques or Algorithms
for Software Aging Prediction Were Used
in the Experiments?
In order to predict aging, data from system resources
are monitored over time, generating a typical dataset
A Systematic Mapping on Software Aging and Rejuvenation Prediction Models in Edge, Fog and Cloud Architectures
937

Table 1: Types of prediction models used in experiments.
Prediction Model Type Paper (Reference)
Regression
a
(Araujo et al., 2011),
(Sudhakar et al., 2014),
(Araujo et al., 2014),
(Di Sanzo et al., 2015),
(Avresky et al., 2017),
(Umesh et al., 2017),
(Liu et al., 2019), (Tang
et al., 2020), (Tan and
Liu, 2021), (Di Sanzo
et al., 2021), (Meng et al.,
2021a), (Meng et al.,
2021b), (Battisti et al.,
2022), (Shi et al., 2023),
(Jia et al., 2023)
Classification (Simeonov and Avresky,
2010), (Yue et al., 2020)
a
Remaining Time To Failure (RTTF) prediction models
regression-based.
as a function of time, that is, a time series. Long
Short-Term Memory (LSTM) neural networks and
their variants are the current state of the art in time
series forecasting.
Models based on the Conv-LSTM network have
demonstrated greater accuracy in predicting TTF and
are cited in Table 2 in works from 2019 onwards.
The Table 3 consolidates the categories shown in
Table 2. There is a quantitative balance between the
techniques used. Classic machine learning algorithms
appear in 6 references, but the most promising recent
state of the art points to hybrid time series analysis
models, based on deep learning (6 references).
The Table 3 consolidates the categories shown in
Table 2.
4.2.3 RQ2 - What Aging Indicators Were
Analyzed in the Experiments?
Most aging is due to computation, network, cache,
RAM and disk contention (Yue et al., 2020), there-
fore, due to their possible exhaustion, these resources
are used as indicators of aging. Typically failures due
to the aging phenomenon occur when free memory
is exhausted, since CPU is a more compressible re-
source. However, among the various aging indicators
pointed out in Table 4, the decline in free RAM mem-
ory (14 references) and CPU consumption (14 refer-
ences) are mentioned more than the others.
Table 2: Algorithms or main techniques that were used in
the experiments.
Technique or Algorithm Paper (Reference)
Time Series Algorithm (Araujo et al., 2011),
(Araujo et al., 2014),
(Umesh et al., 2017),
(Tang et al., 2020),
(Meng et al., 2021a)
Decision Tree (Simeonov and Avresky,
2010), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Di Sanzo et al.,
2021)
CNN+LSTM+Time
Series Algorithm
(Liu et al., 2019), (Bat-
tisti et al., 2022)
CNN+LSTM (Yue et al., 2020), (Tan
and Liu, 2021)
VMD+ARIMA+
BiLSTM
(Shi et al., 2023)
STL+GRU (DGRU) (Jia et al., 2023)
Lasso Regression as a
Predictor
(Di Sanzo et al., 2021)
Support Vector Machine
(SVM)
(Di Sanzo et al., 2021)
ARIMA+ANN (MLP
Feedfoward)
(Meng et al., 2021b)
ANN (MLP
Feedfoward)
(Sudhakar et al., 2014)
Table 3: Type of main approach used in models.
Approach Paper (Reference)
Machine Learning (Simeonov and Avresky,
2010), (Sudhakar et al.,
2014), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Di Sanzo et al.,
2021), (Meng et al.,
2021b)
Deep Learning (Liu et al., 2019), (Yue
et al., 2020), (Tan and
Liu, 2021), (Battisti et al.,
2022), (Shi et al., 2023),
(Jia et al., 2023)
Classic Time Series
Analysis
(Araujo et al., 2011),
(Araujo et al., 2014),
(Umesh et al., 2017),
(Tang et al., 2020),
(Meng et al., 2021a)
ICEIS 2024 - 26th International Conference on Enterprise Information Systems
938

Table 4: Aging indicator resources analyzed in the experi-
ments.
Aging Indicators Paper (Reference)
RAM (Simeonov and Avresky,
2010), (Araujo et al.,
2011), (Araujo et al.,
2014), (Sudhakar et al.,
2014), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Umesh et al.,
2017), (Yue et al., 2020),
(Tang et al., 2020), (Tan
and Liu, 2021), (Di Sanzo
et al., 2021), (Meng et al.,
2021a), (Battisti et al.,
2022), (Jia et al., 2023)
CPU (Simeonov and Avresky,
2010), (Araujo et al.,
2014), (Sudhakar et al.,
2014), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Umesh et al.,
2017), (Liu et al., 2019),
(Yue et al., 2020), (Tan
and Liu, 2021), (Di Sanzo
et al., 2021), (Meng et al.,
2021a), (Meng et al.,
2021b), (Battisti et al.,
2022), (Shi et al., 2023)
Swap (Simeonov and Avresky,
2010), (Araujo et al.,
2011), (Araujo et al.,
2014), (Sudhakar et al.,
2014), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Di Sanzo et al.,
2021), (Battisti et al.,
2022)
Response time for re-
quests or services or
Throughput or Network
transfer rate
(Di Sanzo et al., 2015),
(Yue et al., 2020), (Tang
et al., 2020), (Tan and
Liu, 2021), (Meng et al.,
2021b)
Number of active or
zombie processes or
threads
(Araujo et al., 2011),
(Sudhakar et al., 2014),
(Di Sanzo et al., 2015)
Disk usage (Yue et al., 2020), (Bat-
tisti et al., 2022)
Number of TCP con-
nections
(Sudhakar et al., 2014)
4.2.4 RQ3 - What was the Method Used to
Obtain the Aging Indicator Data in the
Experiments?
Table 5 summarizes the methods for obtaining aging
indicator data used in the experiments. Battisti (Bat-
tisti et al., 2022) reports having used a private dataset
obtained from third parties, generated by an exper-
iment that evaluated the effects of software aging
on a container-based virtualization platform (Oliveira
et al., 2020). Liu (Liu et al., 2019) validated the pre-
diction model with a public dataset obtained from the
Google Cluster Workload.
To validate two experiment scenarios of the soft-
ware aging prediction model, Tan (Tan and Liu, 2021)
used public datasets from Google and Alibaba Cloud
Cluster. The Google Cluster Workload dataset re-
ferred to the resource usage of 1,600 machines over
a 1-month interval, while the Alibaba cluster dataset
referred to the resource usage of 4,000 machines over
8 days.
Shi (Shi et al., 2023) also used a dataset published
by Alibaba in his experiments and Jia (Jia et al., 2023)
also used dataset published by Google Cluster Work-
load. The remaining 12 articles claimed to have used
real data sets collected in the experiments.
Table 5: Method of obtaining data.
Method Paper (Reference)
Monitoring and data
collect data over time
(Simeonov and Avresky,
2010), (Araujo et al.,
2011), (Araujo et al., 2014),
(Sudhakar et al., 2014),
(Di Sanzo et al., 2015),
(Avresky et al., 2017),
(Umesh et al., 2017), (Yue
et al., 2020), (Tang et al.,
2020), (Di Sanzo et al.,
2021), (Meng et al., 2021a),
(Meng et al., 2021b)
Public or private
dataset
(Liu et al., 2019), (Tan and
Liu, 2021), (Battisti et al.,
2022), (Shi et al., 2023), (Jia
et al., 2023)
4.2.5 RQ4 - What Aging Strategies Were
Adopted in the Experiments?
It is proven effective to stress a system or software,
aiming to accelerate the manifestation of bugs or re-
source leaks (Cotroneo et al., 2014). Therefore, it can
be attributed that aging is caused by a large number of
repeated executions that would lead to the accumula-
A Systematic Mapping on Software Aging and Rejuvenation Prediction Models in Edge, Fog and Cloud Architectures
939

tion of system resource leaks (Yue et al., 2020).
Thus, an aging indicator metric may show a clear
upward trend over time, suggesting the occurrence of
the aging phenomenon after a long period of execu-
tion.
Table 6 demonstrates the prevalence of “client re-
quests” to servers over the other types of workload
strategies used in the experiments.
Table 6: Aging strategies analyzed in experiments.
Aging Strategies (Stress
load or Workload)
Paper (Reference)
Client requests or ser-
vice requests
(Di Sanzo et al., 2015),
(Yue et al., 2020), (Meng
et al., 2021a), (Meng
et al., 2021b)
Injection of memory
leaks and unfinished
threads
(Di Sanzo et al., 2015),
(Di Sanzo et al., 2021)
Repeated operations of
instantiating, restarting
and terminating VMs
(Araujo et al., 2011),
(Araujo et al., 2014)
Cloud task requests (Tan and Liu, 2021)
Injection of memory
leaks
(Avresky et al., 2017)
Attaching and detaching
storage volumes
(Araujo et al., 2014)
4.2.6 RQ5 - Which Virtualization Environments
Were Used in Data Acquisition or in the
Aging Experiments?
Virtualization technology can divide the physical
server into multiple virtual machines (VMs) and
cloud-based software services benefit from virtualiza-
tion technology (Tan and Liu, 2021).
On the other hand, containers, in addition to pro-
moting the process of fair and efficient allocation
of physical resources between virtual machines (Yue
et al., 2020), are a form of lightweight virtualization
also widely used to provide services in the cloud, be-
ing subjected to a long cycle of operational life and
intense workload (Oliveira et al., 2020).
These environments are subject to the phe-
nomenon of software aging and Table 7 shows the vir-
tualization environments used by software aging pre-
diction models, regardless of whether they reside in
the cloud.
The virtualization option using VMs appeared in
15 references, while the use of containers was men-
tioned in only 2 references.
Table 7: Virtualization environments used in obtaining the
data or experiments.
Environment Paper (Reference)
Virtual Machines
(VMs)
(Simeonov and Avresky,
2010), (Araujo et al.,
2011), (Araujo et al.,
2014), (Sudhakar et al.,
2014), (Di Sanzo et al.,
2015), (Avresky et al.,
2017), (Umesh et al.,
2017), (Liu et al., 2019),
(Tang et al., 2020), (Tan
and Liu, 2021), (Di Sanzo
et al., 2021), (Meng et al.,
2021a), (Meng et al.,
2021b), (Shi et al., 2023),
(Jia et al., 2023)
Containers (Yue et al., 2020), (Bat-
tisti et al., 2022)
4.2.7 RQ6 - Which Network Architecture has
the Highest Incidence of Studies on
Software Aging Prediction Systems?
Table 8: Network architecture used in the aging experi-
ments or in obtaining the dataset.
Architecture Paper (Reference)
Public Cloud Comput-
ing
(Sudhakar et al., 2014),
(Di Sanzo et al., 2015),
(Avresky et al., 2017),
(Liu et al., 2019), (Yue
et al., 2020), (Tan and
Liu, 2021), (Shi et al.,
2023), (Jia et al., 2023)
Private Cloud Comput-
ing or LAN
a
(Simeonov and Avresky,
2010), (Araujo et al.,
2011), (Araujo et al.,
2014), (Umesh et al.,
2017), (Tang et al., 2020)
Cloud Test Environ-
ment, Simulator or
Framework
(Araujo et al., 2011), (Tan
and Liu, 2021), (Battisti
et al., 2022), (Meng et al.,
2021a), (Meng et al.,
2021b)
Hybrid Cloud (Public
and Private)
(Avresky et al., 2017),
(Di Sanzo et al., 2021)
a
Local Area Network.
Operating environment with continuous and long-
running processing is inherent to cloud computing ar-
chitecture, which is easily prone to producing soft-
ware aging (Liu et al., 2019). Table 8 confirms this
ICEIS 2024 - 26th International Conference on Enterprise Information Systems
940

environment as the target of most data collection or
experiments carried out by researchers.
5 THREATS TO VALIDITY
This section presents the main threats to validity
regarding the conclusions drawn in this secondary
study, namely:
• Although the SLM protocol was initially and im-
partially defined, it is impossible to rule out the
threat to the quality of the articles that were in-
cluded, as the studies were selected without as-
signing scores.
• The consultation carried out in only four scien-
tific literature databases limited the scope of the
search, excluding possible relevant documents ex-
isting in other databases, which can be classified
as a threat to the validity of the work.
• Some inclusion and exclusion criteria may have
eliminated relevant articles and this certainly also
constitutes a threat to the validity of the present
study.
However, due to the rigorous adoption of the SLM
method, it is perfectly possible that other researchers
can replicate the present study and, consequently,
come to confirm similar or equivalent results.
6 CONCLUSIONS
Several techniques and algorithms for software ag-
ing prediction were used in the tabulated experiments.
The specific techniques and algorithms employed var-
ied among the selected articles. The main approaches
proposed in the studies were:
• Time Series Analysis. This technique involves
analyzing historical data on system resource con-
sumption over time to identify patterns and trends
that indicate software aging.
• Machine Learning Algorithms. Various machine
learning algorithms such as regression, decision
trees, support vector machines (SVM) and neural
networks have been used for software aging pre-
diction. These algorithms are trained on historical
data to predict the occurrence of software aging
based on input features.
• Deep Learning Algorithms. Deep learning tech-
niques, particularly convolutional neural networks
and LSTM, have been employed to predict soft-
ware aging. These algorithms can automatically
learn complex patterns and relationships from
large volumes of data, including from transfer
learning, increasing prediction accuracy.
• Statistical Models. Some studies used statisti-
cal models, such as ARIMA (Autoregressive In-
tegrated Moving Average) and others, to analyze
time series data and make predictions about soft-
ware aging. They have been gathered in the cat-
egories Time Series Algorithms and Classic Time
Series Analysis.
• Hybrid Approaches. Other papers have pro-
posed hybrid approaches that combine two or
more techniques or algorithms, with the aim of
chaining and leveraging their respective strengths
in order to improve forecasting accuracy, given
that single aging prediction models typically per-
form poorly and produce less accurate results
(Jia et al., 2023). They were grouped into
CNN+LSTM, CNN+LSTM+Time Series Algo-
rithm, VMD+ARIMA+BiLSTM, STL+GRU cat-
egories. In some articles the Lasso Regression al-
gorithm was not configured as a linear regressive
prediction model. It acted together only in data
regularization for the selection of aging indicators
and, therefore, not as an integral part of a hybrid
prediction model.
It is important to note that the specific techniques
and algorithms used varied between the selected stud-
ies, and more details can be found in the individual
primary articles.
The prediction of software aging is one of the
most important issues in the field of Software Aging
and Rejuvenation (SAR). This article resulted from
an SLM on software aging prediction systems, which
may serve as a reference point for future research on
this highly relevant topic. It provides insights into the
trends and consensus among researchers regarding the
current state of the art.
REFERENCES
Araujo, J., Matos, R., Alves, V., Maciel, P., De Souza, F. V.,
Matias Jr., R., and Trivedi, K. S. (2014). Software ag-
ing in the eucalyptus cloud computing infrastructure:
Characterization and rejuvenation. ACM Journal on
Emerging Technologies in Computing Systems, 10(1).
Cited by: 49.
Araujo, J., Matos, R., Maciel, P., Vieira, F., Matias, R.,
and Trivedi, K. S. (2011). Software rejuvenation in
eucalyptus cloud computing infrastructure: A method
based on time series forecasting and multiple thresh-
olds. In 2011 IEEE Third International Workshop on
Software Aging and Rejuvenation, pages 38–43.
A Systematic Mapping on Software Aging and Rejuvenation Prediction Models in Edge, Fog and Cloud Architectures
941

Avresky, D. R., Pellegrini, A., and Di Sanzo, P. (2017).
Machine learning-based management of cloud appli-
cations in hybrid clouds: A hadoop case study. In
2017 IEEE 16th International Symposium on Network
Computing and Applications (NCA), pages 1–5.
Battisti, F., Silva, A., Pereira, L., Carvalho, T., Araujo, J.,
Choi, E., Nguyen, T. A., and Min, D. (2022). hlstm-
aging: A hybrid lstm model for software aging fore-
cast. Applied Sciences, 12(13).
Cotroneo, D., Natella, R., Pietrantuono, R., and Russo, S.
(2014). A survey of software aging and rejuvenation
studies. J. Emerg. Technol. Comput. Syst., 10(1).
Di Sanzo, P., Avresky, D. R., and Pellegrini, A. (2021).
Autonomic rejuvenation of cloud applications as a
countermeasure to software anomalies. SOFTWARE-
PRACTICE & EXPERIENCE, 51(1, SI):46–71.
Di Sanzo, P., Pellegrini, A., and Avresky, D. R. (2015).
Machine learning for achieving self-* properties and
seamless execution of applications in the cloud. In
2015 IEEE Fourth Symposium on Network Cloud
Computing and Applications (NCCA), pages 51–58.
Grottke, M., Matias, R., and Trivedi, K. S. (2008). The fun-
damentals of software aging. In 2008 IEEE Interna-
tional Conference on Software Reliability Engineering
Workshops (ISSRE Wksp), pages 1–6.
Jia, K., Yu, X., Zhang, C., Hu, W., Zhao, D., and Xiang,
J. (2023). Software aging prediction for cloud ser-
vices using a gate recurrent unit neural network model
based on time series decomposition. IEEE Transac-
tions on Emerging Topics in Computing, 11(3):580–
593.
Kitchenham, B. A., Budgen, D., and Pearl Brereton, O.
(2011). Using mapping studies as the basis for fur-
ther research – a participant-observer case study. In-
formation and Software Technology, 53(6):638–651.
Special Section: Best papers from the APSEC.
Liu, J., Tan, X., and Wang, Y. (2019). Cssap: Software ag-
ing prediction for cloud services based on arima-lstm
hybrid model. In 2019 IEEE International Conference
on Web Services (ICWS), pages 283–290.
Meng, H., Shi, Y., Qu, Y., Li, J., and Liu, J. (2021a). Arima-
based aging prediction method for cloud server sys-
tem. volume 1043.
Meng, H., Tong, X., Shi, Y., Zhu, L., Feng, K., and Hei, X.
(2021b). Cloud server aging prediction method based
on hybrid model of auto-regressive integrated moving
average and recurrent neural network;. Tongxin Xue-
bao/Journal on Communications, 42(1):163 – 171.
Oliveira, F., Araujo, J., Matos, R., Lins, L., Rodrigues, A.,
and Maciel, P. (2020). Experimental evaluation of
software aging effects in a container-based virtualiza-
tion platform. In 2020 IEEE International Conference
on Systems, Man, and Cybernetics (SMC), pages 414–
419.
Oliveira, F., Araujo, J., Matos, R., and Maciel, P. (2021).
Software aging in container-based virtualization:an
experimental analysis on docker platform. In 2021
16th Iberian Conference on Information Systems and
Technologies (CISTI), pages 1–7.
Petersen, K., Vakkalanka, S., and Kuzniarz, L. (2015).
Guidelines for conducting systematic mapping stud-
ies in software engineering: An update. Information
and Software Technology, 64:1–18.
Shi, F., Yuan, Z., Wang, M., and Cui, J. (2023). Research
on cloud platform software aging prediction method
based on vmd-arima-bilstm combined model. Inte-
grated Ferroelectrics, 237(1):297 – 309.
Simeonov, D. and Avresky, D. R. (2010). Proactive soft-
ware rejuvenation based on machine learning tech-
niques. In Avresky, D., Diaz, M., Bode, A., Ciciani,
B., and Dekel, E., editors, CLOUD COMPUTING,
volume 34 of Lecture Notes of the Institute for Com-
puter Sciences, Social Informatics, and Telecommuni-
cations Engineering, pages 186–200.
Sudhakar, C., Shah, I., and Ramesh, T. (2014). Software
rejuvenation in cloud systems using neural networks.
In 2014 International Conference on Parallel, Dis-
tributed and Grid Computing, pages 230–233.
Tan, X. and Liu, J. (2021). Aclm:software aging prediction
of virtual machine monitor based on attention mech-
anism of cnn-lstm model. In 2021 IEEE 21st Inter-
national Conference on Software Quality, Reliability
and Security (QRS), pages 759–767.
Tang, M., Zhang, P., Sun, H., and Zhang, L. (2020).
A task execution framework based on aging indica-
tor and sarimi. In 2020 IEEE Intl Conf on Paral-
lel & Distributed Processing with Applications, Big
Data & Cloud Computing, Sustainable Computing
& Communications, Social Computing & Networking
(ISPA/BDCloud/SocialCom/SustainCom), pages 18–
25.
Umesh, I., Srinivasan, G., and Torquato, M. (2017). Soft-
ware aging forecasting using time series model. In-
donesian Journal of Electrical Engineering and Com-
puter Science, 8(3):589 – 596. Cited by: 6; All Open
Access, Green Open Access.
Wang, S. and Liu, J. (2020). Harrd: Real-time software
rejuvenation decision based on hierarchical analysis
under weibull distribution. In 2020 IEEE 20th Inter-
national Conference on Software Quality, Reliability
and Security (QRS), pages 83–90.
Yue, J., Wu, X., and Xue, Y. (2020). Microservice aging
and rejuvenation. In 2020 World Conference on Com-
puting and Communication Technologies (WCCCT),
pages 1–5.
ICEIS 2024 - 26th International Conference on Enterprise Information Systems
942