The current literature points out that the scarcity of architectural documentation predominates in free software projects. In this sense, several techniques have been proposed to retrieve information about the architecture, and most of them focus on structural aspects from the source code, an element that accompanies the evolution of the software product.
Problem. However, these techniques do not retrieve other information about the architecture (nor its evolution), such as the objectives, restrictions and motivations, non-functional requirements, data access or communication mechanisms used in the project.
Objective. The purpose of this work is to characterize architectural changes in Free Software projects, presenting the impacts caused by them in the project's source code in comparison with other types of changes in which the architecture has not been modified. The hypotheses are that the commit messages record information about changes in the project architecture and this type of change impacts the project code differently compared to changes where the architecture has not been modified.
Methodology. Initially, an empirical study was produced, from the mining of repositories, in which architectural traces were identified in records of information about changes contained in commit messages. From the data collected, an exploratory study was carried out that characterized such changes about when the project's evolution occurred, who are its authors, which architectural topics are modified, in addition to the characteristics of these messages and the group of modules affected by the change . After this study, a second empirical study was conducted, involving the extraction of metrics from source code of the versions of the project identified as having undergone modifications in their architecture. Finally, a study was carried out to evaluate the results obtained, considering the opinions of researchers in the field of Architecture and Software Engineering, as well as of developers regarding the characterization obtained in the face of architectural changes.
Results. Among the discoveries made, it can be noted that periods of release of new versions are concentrators of architectural changes, the main collaborators of the project are also the collaborators who most modify the architecture and all the architectural topics of the project are subject to modifications, however in different scales. Other important aspects of architectural changes are that (i) the size of the code tends to increase; (ii) the complexity and cohesion of the code tends to remain; and (iii) the verified coupling increased to two metrics and in three others it was maintained, when compared to changes in the design where the architecture was not reached. Finally, during the third study of this thesis, developers and researchers confirmed the hypothesis that architectural documentation is scarce in Free Software projects, and the main way to retrieve information about architecture, until now, is by analyzing the source code of several shapes. Most research participants agree with the results obtained by this study in relation to the variation of code metrics, both as being generalizable and verifiable in the projects where they research and develop. Some developers showed slight disagreements as to the variations of metrics obtained. Regarding architectural features as sources of information about architecture, most developers and researchers claimed to be a promising source of information but made the reservation that an ideal scenario would be to combine them with other information from the project.

Competition for network resources needs to be arbitrated according to the characteristics of each application or service to provide a satisfactory, transparent and cost-effective manner for users. Meeting resource requests dynamically, that is, without prior allocation (scheduling) of resources for applications and services can allow sharing of them for better use of the network. Due to the heterogeneity of user profiles, applications and services, dynamic response to resource requests tends to generate dynamic competition for resources across networks as a whole. Bandwidth allocation models have the attributes that allow you to define application classes and control the distribution of resources between classes intuitively. These models can be changed and / or reconfigured to optimize resource utilization to evolve their behaviors in line with the traffic profile and communication and network quality requirements. Defining behavior in resource arbitration that reflects better efficiency of network quality and communication requirements defined in a given traffic profile is a complex task. This complexity can make human intervention a point of failure. Generally speaking, cognitive management systems are designed to handle complex tasks where human intervention can be a point of failure. These systems are able to autonomously configure themselves in response to changes or failures in accordance with business policies specified by administrators. This thesis proposes the cognitive and dynamic orchestration of BAM-based resource allocation (bandwidth) for MPLS / DSTE networks. To this end, they propose a new GBAM band allocation model that generalizes the behaviors of classic models and allows new strategy combinations to be configurable at runtime; and a cognitive framework (Cognitive BAM) that dynamically learns and orchestrates GBAM behavior in line with the traffic profile and communication and quality requirements set for the network.

This project proposes new Orchestration strategies based on Network and Application tiers awareness to accommodate the growing needs for Cloud Computing and its application. Commonly used orchestration strategies use CPU, RAM, Network, and storage attributes to allocate resources. The prototype of Network and Application Aware Orchestration (NAAO) is capable of orchestrating resources using, not only CPU, RAM and Network, but it also considers application tiers that are highly inter-dependent to each other. Those tiers are orchestrated as close as possible to provide the lowest latency between them. Three different strategies were developed: Aggressive, Aggregate, and Balanced. Each strategy can be applicable to meet different application latency, and response time SLAs. For each strategy, besides latency, VM migration and VM allocation time were measured. Each strategy is compared to commonly used orchestration strategy, and it is also compared against each other to understand the most suitable use case for each one. Aggregate and Aggressive strategies results can guarantee the lowest latency possible due to its behavior, whereas Balance has higher latency compared to these two, but very low latency when compared to commonly used orchestration. For Balanced and Aggregate, VM migration for 2-tier application occurs on average 14.51% and 22.82%, respectively, and it increases to 24.18% and 40.40% when 3-tier applications are used. On average, all three strategies have 4.04x higher VM allocation times when compared to commonly used orchestration. For all use cases tested in this dissertation, NAAO is able to achieve very low latency for highly inter-dependent applications (2 and 3 tiers), however it has impacts on VM migration and VM allocation time, which are tradeoffs of the proposed strategies.

Traceability in Software Product Lines (SPL) refers to the interrelation of software artifacts through required links to answer specific questions about families of products and their underlying development processes. Traceability becomes even more complex in SPLs if compared to single system development due to additional issues posed by the need for variability management in all development stages. Such variability should be systematically traced to cope with a consistent product line evolution. This thesis initially generated a systematic mapping study evidence by analyzing 62 SPL traceability primary studies. A set of aspects for such research field were investigated: main goals, strategies, application domains, research intensity, challenges, research rigor, and industrial relevance. The mapping study findings indicated a set of research gaps as the one addressed by this work, the lack of SPL traceability recovery (semi-) automatic techniques. Although there is a set of studies that perform traceability recovery using information retrieval methods in a single system development setting, none of them have targeted the SPL application domain. Thus, another contribution of this work concerned the evaluation of IR-based proposals as traceability recovery instruments since they represent a successful alternative for single system scenarios. The evaluation performed in a controlled-experiment fashion applied five different IR methods (namely classic vector, extended boolean, latent semantic indexing, neural network and BM25) into 41 SPL projects aiming at recovering feature-to-code trace links. The experimental results identified limitations of IR proposals related to low recovery precision rates and the inability to collect information related to variability. Considering the limitations of current research, this thesis proposes trace2vary, a traceability recovery algorithm that creates relevant feature-code trace links with embodied variability information. The evaluation results from the application of the proposal in 16 real-world projects with a total of, approximately, 61 million lines of code, indicate that trace2vary presents improvements in the traces recovery precision and the ability to provide variability knowledge at both feature and source code levels.