Article

October 27, 2026

Development of a System for Migrating Data from Relational Databases to NoSQL

Lucas José Martins de Lara; Alexandre Leite Rangel

DOI: 10.22167/2675-6528-202601127

Article prepared by the ResumeAI tool, an artificial intelligence solution developed by the Pecege Institute focused on synthesis and writing.

Summary

The advancement of database technologies has driven the need for migration between relational and NoSQL systems, due to differences in their data models and scalability requirements. An automated system was developed to convert data from MySQL and PostgreSQL to MongoDB and Cassandra, aiming to preserve the integrity and structure of information. The experimental research used the Python language and various libraries to analyze schemas and transform data into three distinct models: normalized relational, sparse relational, and Entity-Attribute-Value (EAV), in a containerized environment with Docker. It was observed that MongoDB showed higher speed for traditional relational models, whether fully populated or sparse. However, it faced a critical performance bottleneck in the EAV model, where the computational cost of data pivoting resulted in significantly high execution times. In contrast, Cassandra proved to be more efficient for the EAV model, in terms of both ingestion speed and disk storage economy. It was concluded that there is no universal solution for NoSQL migration, and the choice of the target technology should be guided by the structural nature of the source data. MongoDB proved agile for normalized tables, while Cassandra established itself as the ideal option for scenarios of high flexibility and sparsity. The functional tool and the comparative metrics generated contributed to strategic decision-making in data infrastructure modernization projects.

Keywords: Automation; Cassandra; Scalability; MongoDB; Data models.

1. Introduction

Data management has experienced a remarkable evolution over the past few decades, driven by the exponential growth in the volume, variety, and velocity of information. Traditionally, relational database management systems (RDBMS) have been the backbone for storing structured data, offering a robust model based on interconnected tables and guaranteeing ACID properties (Atomicity, Consistency, Isolation, and Durability) (Elmasri and Navathe, 2020). However, the growing demand for applications with high scalability, schema flexibility, and the ability to process large volumes of unstructured or semi-structured data has led to the emergence and popularization of NoSQL (Not only SQL) databases (Sethi and Panda, 2024). Although RDBMS still dominate the market, accounting for over seventy percent of use cases (DB-Engines, 2025), the need to modernize data infrastructures and leverage the advantages of NoSQL paradigms has generated significant demand for efficient migration solutions (Sadalage and Fowler, 2013).

The transition of data between relational and NoSQL environments, however, presents complex challenges, mainly due to fundamental differences in their data models and operational philosophies. Relational databases operate with fixed, normalized schemas, optimized for integrity and complex queries via SQL. In contrast, NoSQL databases adopt flexible or schema-less schemas, prioritizing horizontal scalability and availability, with models ranging from document, graph, key-value, to wide-column stores (Vilaça et al., 2013). This structural incompatibility, often referred to as “schema impedance mismatch”, makes migration an arduous task, requiring not only data transformation but also application adaptation and the preservation of informational integrity (Ghotiya, 2017). Optimizing this process requires tools that automate schema analysis, data conversion, and application adaptation, minimizing manual interventions and potential errors (Sethi and Panda, 2024). Aspects such as performance and scalability are crucial, and handling large volumes of data in heterogeneous systems poses significant challenges in organizing and translating queries (Elmasri and Navathe, 2020; Malik et al., 2020).

To address the complexity of migration, this study explores data transformation in three distinct models: normalized relational, sparse relational, and Entity-Attribute-Value (EAV). The normalized model represents the traditional relational structure, with well-defined tables and primary and foreign keys to ensure referential integrity. The sparse model introduces data variability, simulating real-world scenarios where many fields may be empty or null, testing the resilience of the migration system to heterogeneous and incomplete schemas. Finally, the EAV model is a flexible architecture that stores information as sets of attribute-value pairs, rather than fixed columns, being particularly useful for varied and complex data, such as medical records, where each item may have different properties (Dinu and Nadkarni, 2006). The transposition of EAV data to NoSQL paradigms is especially challenging, as it requires pivoting and restructuring operations to consolidate sparse attributes into documents or columns, which can significantly impact performance and storage consumption.

Given the scenario of constant technological evolution and the growing need for flexibility in data management, the migration between relational and NoSQL databases has become a strategic imperative for many organizations. The absence of a universal solution and the inherent complexity of this transition justify the need for tools that optimize and standardize the process, ensuring data integrity and operational efficiency. Thus, the present work is justified by its contribution to the modernization of data infrastructures, offering an automated system for migrating data from relational databases (MySQL and PostgreSQL) to NoSQL databases (MongoDB and Cassandra), with the objective of analyzing performance and storage efficiency in different data models (normalized, sparse, and EAV) and providing comparative metrics to aid in strategic decision-making.

2. Material and Methods

The present work is characterized as an Experimental Research, according to the classification proposed by Cajueiro (2015). This methodological approach was selected because it involves the implementation and testing of new technologies and algorithms, specifically a system for data migration. Experimental research allowed for the systematic evaluation of hypotheses related to performance and storage efficiency in different migration scenarios, providing an empirical basis for the study’s conclusions.

The execution of the migration process and the performance tests were conducted in a local machine environment. This environment consisted of a 13th generation Intel Core i7 processor, equipped with 16 GB of DDR5 RAM, a 1 TB SSD storage, and an Nvidia GeForce RTX 4060 GDDR6 graphics card. The development and testing infrastructure was configured using Docker containers (version 27.2.0), which ensured the isolation of each database technology and facilitated the reproducibility of the experiments conducted.

The study focused on data migration between relational and NoSQL databases, using MySQL (version 8.4.5) and PostgreSQL (version 15.13.1) as relational sources, and MongoDB (version 6.0.23) and Cassandra (version 4.1.9) as NoSQL destinations. The unit of analysis consisted of data transformation and transfer processes, evaluating their performance and storage efficiency across three distinct data models: normalized, sparse, and Entity-Attribute-Value (EAV), aiming for a comprehensive comparative analysis.

To simulate a real environment with a large volume of records, synthetic user registration data were generated. A Python class, named FakeData, was used, which employed the Faker library (version 37.3.0) to create this data. The data were structured into three distinct models, each with 100,000 records, and initially inserted into the MySQL and PostgreSQL relational databases, serving as the source population for subsequent migration tests.

In the first data model, the relational structure was normalized, comprising four interconnected tables: `users`, `addresses`, `social_networks`, and `credit_cards`. All columns in these tables were fully populated, with the `users` table containing seven columns and using an ID as the primary key. The other tables employed foreign keys to maintain referential integrity with the main table. The second model maintained the same topology but introduced sparsity, with columns such as `age` and `profession` in the `users` table having a 15% and 30% probability of being empty, respectively. Other tables had nullity rates ranging between 5% and 40%.

The third data model adopted the Entity-Attribute-Value (EAV) architecture, consolidating all information into a single table with four columns: `id`, `entity`, `attribute`, and `value`. User CPF was used as the logical identifier in the `entity` column, while the `attribute` column dynamically mapped seventeen biographical and complementary fields. To increase complexity, the same probabilities of attribute non-existence observed in the sparse model were applied, challenging the tool to transpose this flexible and sparse structure to MongoDB’s document models and Cassandra’s columns.

Metadata extraction and structural analysis of the source databases were performed using the RelationalDatabaseInspector class. This class utilized the `inspect` functionality of the SQLAlchemy library (version 2.0.49) to automatically map referential integrity dependencies, such as Primary Keys (PK) and Foreign Keys (FK). In addition to classic relational inspection, the tool implemented logic to detect structures based on the EAV model, identifying patterns in column triplets (`entity`, `attribute`, `value`), allowing the application of the most suitable data transformation strategy for each case.

The migration to MongoDB was operationalized by the RelationalToMongoMigrator class, which worked in conjunction with RelationalDatabaseInspector to determine the persistence strategy. For the normalized and sparse models, the class used the mapping of primary and foreign keys to perform the join of data from the tables, consolidating them into unique documents, where the original Primary Key was mapped directly to the MongoDB `_id` field. In the case of the EAV model, the tool processed the table by grouping the attributes of each entity through a pivoting operation with the Pandas library, defining the CPF as the index and natural key of the destination document, given the absence of a typical sequential numeric identifier.

For the migration to Cassandra, the operationalization was conducted by the RelationalToCassandraMigrator class, which also worked together with RelationalDatabaseInspector to map the source infrastructure. The process began with connecting to the relational database via SQLAlchemy, followed by data extraction and the dynamic creation of keyspaces and tables in Cassandra, based on automatically identified fields. In the first two models (normalized and sparse), the inspection of primary and foreign key relationships allowed the execution of queries with `joins` to enable this flow, implementing a column disambiguation logic to avoid naming conflicts between distinct tables during persistence in the column-oriented model.

In the third model, focused on the EAV architecture, the system consolidated data by entity through a pivoting operation with the Pandas library, transforming sparse attributes into structured columns before insertion. In this context, the values contained in the `entity` column were defined as the Primary Key in Cassandra, ensuring the distribution and efficient access to records in the wide-column model. Throughout the entire procedure, fine-tuning of data types and the definition of partition keys that respected the best practices for scalability of the target database were performed.

The infrastructure and volumetry analysis was developed through the DBStorageAnalyzer class, which used the `docker` library for Python to interact with the containerized environment. The monitoring of the size occupied by the four databases (MySQL, PostgreSQL, MongoDB, and Cassandra) was performed by executing the command “du -sb {path}” in each container. The “path” parameter pointed to the specific data storage directory of each technology, allowing for precise extraction of disk consumption in bytes to support the comparative analysis.

To automate the testing of different migration routes, the DatabaseMigrationTester class was created. This class centralized and executed the entire process of data generation, migration, and visualization of results. It allowed for standardized testing of the four migration routes: from MySQL to MongoDB, from MySQL to Cassandra, from PostgreSQL to MongoDB, and from PostgreSQL to Cassandra, ensuring the correct execution of each step of the flow. Time counting was incorporated using the `perf_counter()` function from the `time` library, recording the start and end of each migration to evaluate process performance.

The collected data, regarding execution times and disk storage consumption, were subjected to a comparative analysis. For each of the three data models (normalized, sparse, and EAV) and for each of the four migration routes, ten test cycles were recorded. The averages of these cycles were calculated for each scenario, allowing for the evaluation of the temporal and volumetric efficiency of the different destination technologies. This approach aimed to identify performance and resource consumption patterns, providing metrics to assist in strategic decision-making regarding the suitability of each NoSQL paradigm.

3. Results and Discussion

The successful implementation of the `RelationalToMongoMigrator` and `RelationalToCassandraMigrator` classes demonstrated the technical feasibility of an automated system for data migration between relational databases, such as MySQL and PostgreSQL, and NoSQL databases, specifically MongoDB and Cassandra. The tool’s full functionality was ensured by its integration with the `RelationalDatabaseInspector` class, which performed automated metadata analysis. This crucial step allowed for the precise identification of primary keys (PK), foreign keys (FK), and, innovatively, the detection of structures based on the Entity-Attribute-Value (EAV) model. The ability to inspect and understand the source data structure was fundamental for the correct execution of all phases of the migration process, from initial connection and table reading to logical data transformation and insertion into the target databases. The automation of these inspection and transformation processes is a significant advancement, as, according to Ghotiya (2017), migration between different database paradigms is a complex task, mainly due to disparities in data models. The developed tool mitigates this complexity by standardizing and optimizing the steps, reducing the need for manual intervention and the risk of errors.

The migration operations were orchestrated by the `DatabaseMigrationTester` class, which centralized and standardized the complete flow for the four defined migration routes: MySQL to MongoDB, PostgreSQL to MongoDB, MySQL to Cassandra, and PostgreSQL to Cassandra. This systematic approach ensured the consistent and reproducible execution of experiments, a pillar of experimental research as per Cajueiro (2015). The system’s modular architecture, with specialized components for each phase of the process, is illustrated in Figure 1, which details how the `DatabaseMigration` class acts as a central orchestrator, delegating specific tasks to the inspection, data generation, and migration classes.

Source: Elaborated by the author (2026)

Figure 1. Class Diagram

The visual representation in Figure 1 highlights the interconnection between modules, such as `DatabaseMigration` utilizing `FakeData` for synthetic data generation, `DatabaseInspector` for schema analysis, and specific migrators (`RelationalToMongoMigrator` and `RelationalToCassandraMigrator`) for executing data transfer. This modular structure not only facilitates understanding of the workflow but also promotes system maintainability and scalability, allowing new modules or database technologies to be integrated with relative ease in the future. The use of Docker containers (version 27.2.0) to isolate each database technology, as detailed in the methodology, reinforced the reproducibility of experiments, a crucial aspect for validating results and for the credibility of conclusions (Sethi & Panda, 2024). The ability to successfully test the twelve planned combinations (four migration routes for each of the three data models), validating data integrity through direct queries on the destinations, underscores the robustness of the proposed solution.

Regarding the temporal performance of migrations, the results revealed a significant divergence between the target databases, depending on the source data structure. For models 1 (normalized relational) and 2 (sparse relational), migration routes directed to MongoDB showed, on average, greater temporal efficiency compared to Cassandra. For example, the migration from MySQL to MongoDB for the normalized model (Table 1) had an average of 53.01 seconds, while for Cassandra it was 79.64 seconds. Similarly, for PostgreSQL, the migration to MongoDB was 35.08 seconds, versus 62.03 seconds for Cassandra. This superior performance of MongoDB can be attributed to its document-oriented architecture, which adapts well to consolidating relational data into nested structures. The `RelationalToMongoMigrator` class used the mapping of primary and foreign keys to perform data joining from tables, consolidating them into single documents, where the original primary key was mapped directly to MongoDB’s `_id` field. This approach minimizes transformation and insertion overhead, as MongoDB’s document model is intrinsically flexible and allows for efficient representation of complex and hierarchical data (Elmasri & Navathe, 2020). MongoDB’s schema flexibility (Vilaça et al., 2013) allows data to be inserted without the need for a rigid predefined schema, which speeds up the migration process for structured data that can be denormalized into documents.

In contrast, the migration to Cassandra for the normalized and sparse models, although successful, registered higher write latencies. The `RelationalToCassandraMigrator` class performed the dynamic creation of keyspaces and tables in Cassandra, based on automatically identified fields. For the first two models, the inspection of primary and foreign key relationships allowed the execution of queries with `joins` to enable this flow, implementing a column disambiguation logic to avoid naming conflicts. Cassandra’s wide-column model, while optimized for horizontal scalability and high availability, can introduce a greater overhead for the initial ingestion of relational data that requires complex transformations to fit its data model. The need to define partition keys and group data in an optimized way for distributed queries can add complexity and time to the initial migration process, especially when compared to the simplicity of inserting documents into MongoDB for already consolidated data. This performance difference suggests that, for workloads originating from traditional relational schemas, MongoDB offers greater agility in processing, regardless of data density, as observed in the normalized and sparse models.

However, scenario 3, based on the Entity-Attribute-Value (EAV) architecture, revealed a complete inversion in the performance profile. As presented in Table 1, Cassandra demonstrated a notable superiority in ingesting this data, while MongoDB presented a critical bottleneck, with execution times exceeding 15 minutes in all tests performed.

Table 1. Times in seconds for migration of 100,000 rows to model 3

Test

MySQL → MongoDB Runtime

PostgreSQL → MongoDB Runtime

MySQL Runtime → Cassandra

PostgreSQL → Cassandra Runtime

1

1671,86

945,39

193,86

109,65

2

1641,45

957,39

192,72

108,57

3

1612,43

932,76

191,07

107,29

4

1705,81

968,14

198,33

113,46

5

1658,94

912,47

186,22

105,13

6

1594,27

985,33

204,65

112,71

7

1722,56

944,81

196,38

109,54

8

1637,12

1004,59

189,74

104,28

9

1674,38

926,15

201,47

111,06

10

1691,02

957,38

193,58

108,67

Average(s)

1660,98

953,44

194,80

109,04

Source: Original research results

Table 1 clearly illustrates this disparity: migrating from MySQL to MongoDB for the EAV model took an average of 1660.98 seconds (approximately 27.68 minutes), while for Cassandra, the same migration was completed in 194.80 seconds (approximately 3.25 minutes). Similarly, migrating from PostgreSQL to MongoDB took 953.44 seconds (approximately 15.89 minutes), compared to 109.04 seconds (approximately 1.82 minutes) for Cassandra. This substantial discrepancy can be attributed to the intrinsic characteristics of the EAV model and how each NoSQL database handles it. The EAV model, as described by Dinu and Nadkarni (2006), stores information as sets of attribute-value pairs, which provides great flexibility but can be challenging for systems not designed for such a structure. For MongoDB, the tool processed the EAV table by grouping the attributes of each entity using a pivoting operation with the Pandas library. This pivoting operation, which transforms rows into columns to consolidate an entity’s attributes into a single document, is computationally intensive. Reconstructing complex documents from a sparse EAV structure generates significant overhead, highlighting performance limitations for this specific migration paradigm, even in controlled environments.

In contrast, Cassandra, with its wide-column architecture, proved exceptionally efficient for the EAV model. The system performed data consolidation by entity through a pivoting operation with the Pandas library before insertion, but the nature of Cassandra’s wide-column model allows sparse attributes to be stored naturally and efficiently. The values contained in the `entity` column were defined as the Primary Key in Cassandra, ensuring efficient distribution and access to records. Cassandra’s ability to handle a large number of dynamic columns, where each attribute can be a column, aligns perfectly with the flexibility of EAV, without the need for complex transformations that penalize performance. This finding corroborates literature highlighting the suitability of wide-column databases for sparse and flexible data, where scalability and write efficiency are paramount (Heine et al., 2018). Cassandra’s robustness in handling the complexity and sparsity of the EAV model, without incurring performance bottlenecks, positions it as an ideal choice for highly variable data scenarios.

The infrastructure and volumetrics analysis, performed by the `DBStorageAnalyzer` class, provided crucial insights into the disk storage efficiency of different databases. Tables 4, 5, and 6 presented measurements of the space occupied by tables or collections within containers, allowing for a comparative analysis of the volumetric efficiency of each technology. The data demonstrated that storage efficiency varies drastically according to the applied logical structure. In the scenario of normalized tables (Model 1), as detailed in Table 4, the relational databases MySQL and PostgreSQL showed the lowest space consumption, remaining in the range of 10 MB to 15 MB. For example, the average storage for MySQL was 11.78 MB, and for PostgreSQL, 14.63 MB. In contrast, MongoDB recorded the highest values, frequently exceeding 25 MB, with an average of 24.99 MB for MySQL data and 19.89 MB for PostgreSQL data. This difference in occupation suggests that, for fixed and standardized data structures, the normalized table disk storage model is more compact than MongoDB’s document-based approach. The inherent denormalization in consolidating data into documents can lead to information duplication, increasing the total storage volume, a common trade-off in NoSQL databases to optimize read performance (Sadalage & Fowler, 2013).

In the case of sparse data (Model 2), as shown in Table 5, it is observed that MySQL maintained a consumption stability close to the original scenario, oscillating slightly between 9.87 MB and 11.56 MB, with an average of 10.72 MB. PostgreSQL also maintained stable consumption, with an average of 13.85 MB. MongoDB, on the other hand, continued to present constant and high consumption, remaining the option that consumes the most disk space for flows originating from both MySQL (average of 24.84 MB) and PostgreSQL (average of 18.26 MB). Cassandra, in turn, proved to be more efficient than MongoDB for sparse data, with averages of 13.91 MB and 13.53 MB, respectively. In summary, the data indicate that the space gain in using NoSQL is conditioned by the transformation of the data architecture, and not just by the direct migration of existing tables. Cassandra’s storage efficiency for sparse data can be attributed to its ability to store only columns that contain values, avoiding the overhead of null fields that could be represented in MongoDB documents.

Table 2. Storage of Tables and Collections for model 3

Test

MySQL (MB)

MongoDB (MB)

Cassandra (MB)

PostgreSQL (MB)

MongoDB (MB)

Cassandr a (MB)

1

89,77

35,05

17,45

131,7

22,71

11,01

2

89,77

33,42

17,48

131,69

23,24

8,93

3

87,46

32,17

18,12

135,11

21,93

9,46

4

92,34

36,08

16,97

128,49

24,16

10,53

5

84,63

34,81

17,29

140,22

22,54

8,84

6

94,18

33,69

18,03

122,86

23,81

11,42

7

88,12

35,42

17,66

137,54

21,47

9,82

8

91,26

32,94

16,84

126,38

24,03

10,19

9

86,59

36,21

17,92

139,07

22,18

9,13

10

93,71

34,16

18,34

124,19

23,57

10,81

Average(s)

89,76

34,40

17,61

131,70

22,96

10,01

Source: Original research results

In contrast, the EAV (Model 3) table scenario, as shown in Table 2, revealed a complete inversion of performance in terms of storage. PostgreSQL, which occupied approximately 14 MB in a normalized relational structure, jumped to an average of 131.70 MB in EAV, representing the largest data volume recorded in the entire study. This drastic increase in space consumption in relational databases for the EAV model is a direct consequence of how this data is stored: each attribute-value for an entity requires a new row, resulting in a proliferation of records and the repetition of entity and attribute identifiers, which is highly inefficient for storage (Dinu & Nadkarni, 2006).

Meanwhile, Cassandra demonstrated high efficiency in this same workflow, reducing storage to values between 8.84 MB and 11.42 MB, with an average of 10.01 MB. Cassandra’s superior performance for EAV data is a testament to its wide-column architecture, which allows for efficient storage of sparse attributes as dynamic columns, without the overhead of repeated rows or explicit null fields. MongoDB, although it presented a high computational cost for pivoting, was still more efficient in terms of storage for the EAV model than PostgreSQL, with an average of 34.40 MB for MySQL data and 22.96 MB for PostgreSQL data. The data indicate that NoSQL databases, especially Cassandra, handle the denormalization of large data volumes more economically than traditional relational databases, particularly when the data structure is inherently flexible and sparse like EAV. This finding is of utmost importance for projects involving large volumes of heterogeneous data, where storage optimization can result in substantial infrastructure savings.

The `RelationalDatabaseInspector`’s ability to detect and differentiate between normalized tables and EAV structures by analyzing patterns in column triplets (`entity`, `attribute`, `value`) was a crucial differentiator. This pre-migration intelligence allowed the system to apply the most suitable data transformation strategy for each case, which is vital for the integrity and performance of the migration (Malik et al., 2020). Without this automatic detection, migrating EAV data to a NoSQL database could result in suboptimal structures or require considerable manual effort for schema re-engineering, which contradicts the goal of automation and efficiency. The proposed tool, therefore, not only transfers data but also interprets and adapts the transformation logic based on the structural semantics of the source data, an aspect that Ghotiya (2017) and Sethi and Panda (2024) highlight as one of the biggest challenges in cross-paradigm migrations.

The automation of tests through the `DatabaseMigrationTester` class, which incorporated time counting using `perf_counter()` from the `time` library, allowed for a precise evaluation of the processes’ performance. This reusable, expandable, and measurable structure is fundamental for conducting new tests and experiments, ensuring that the tool can be adapted and improved to handle even more complex migration scenarios or different database technologies. The standardization of tests and the collection of performance and storage metrics are essential for providing a solid empirical basis for strategic decision-making in data infrastructure modernization projects.

In summary, the experimental results revealed that migration efficiency is intrinsically linked to the compatibility between the source structure and the target database storage model. It was observed that MongoDB exhibits superior processing speed performance for traditional relational models, whether fully populated or sparse. This is due to its ability to consolidate data into documents, which aligns well with denormalization for read optimization. However, this performance advantage may come with higher disk consumption compared to the original relational databases.

On the other hand, the scenario based on the Entity-Attribute-Value (EAV) architecture showed a critical performance bottleneck in MongoDB, where the computational cost of data pivoting resulted in significantly high execution times. In contrast, Cassandra proved to be the most resilient and efficient technology for handling the EAV model, both in terms of ingestion speed and disk storage economy. Its wide-column architecture is naturally suited for sparse and flexible data, avoiding the overhead of complex transformations and wasted space. Relational databases, such as PostgreSQL, suffered severe volumetric degradation when representing data in EAV format, reaching the peak consumption recorded in the study, while Cassandra maintained reduced and stable disk consumption.

It is concluded, therefore, that there is no universal solution for NoSQL migration. The choice of the target technology should be guided by the structural nature of the source data and by performance and storage requirements. For normalized and sparse tables with fixed structures, MongoDB offers migration agility, albeit with potentially higher disk consumption than the relational database model. For scenarios requiring high flexibility and sparsity, such as EAV, Cassandra consolidates itself as the ideal option to ensure scalability and efficiency, both in migration time and storage usage. This work therefore contributes with a functional tool and comparative metrics that assist in strategic decision-making in data infrastructure modernization projects, allowing data engineers and system architects to make informed choices that optimize both performance and operational costs.

4. Conclusion

It is concluded that the objective was achieved, with the development of an automated system for migrating data from relational databases (MySQL and PostgreSQL) to NoSQL databases (MongoDB and Cassandra). The system allowed for the analysis of performance and storage efficiency across different data models (normalized, sparse, and EAV), providing crucial comparative metrics for strategic decision-making. The results demonstrated that migration efficiency is intrinsically linked to the compatibility between the source structure and the destination database’s storage model. MongoDB proved more agile for traditional relational models (normalized and sparse), albeit with higher disk consumption. However, for the EAV model, MongoDB presented a critical performance bottleneck due to the computational cost of data pivoting, highlighting a significant limitation for this specific migration paradigm.

In contrast, Cassandra proved to be the most efficient technology for the EAV model, both in ingestion speed and storage economy, given its wide-column architecture. This research reinforces that there is no universal solution for NoSQL migration, with the choice of the target technology depending on the structural nature of the data and the performance and storage requirements. Limitations of the study include the focus on a specific set of databases and data models, and a volume of 100,000 records. For future studies, it is recommended to explore eventual consistency mechanisms, include other types of NoSQL databases, and conduct tests with data volumes on the terabyte scale, in order to validate the scalability of the solution in production scenarios.

Bibliographic References

autor, 2026 [Referência completa não encontrada no documento original]

Cajueiro, Roberta Liana Pimentel. Manual para Elaboração de Trabalhos Acadêmicos: guia prático do estudante. Rio de Janeiro: Saraiva, 2015. 159 р.

DB-Engines. 2025. Popularidade de DBMS dividida por modelo de banco de dados. Disponível em: <https://db-engines.com/en/ranking_categories>. Acesso em: 24 mai. 2025.

Dinu, V.; Nadkarni, P. 2007. Guidelines for the effective use of entity-attribute-value modeling for biomedical databases. International journal of medical informatics, 76 (11-12), 769-779.

Elmasri, R.; Navathe, S. B.; Halpin, T. 2020. Chapter 1 Databases and Database Users. p. 3-29. In: Elmasri, R.; Navathe, S. B.; Halpin, T. Fundamentals of Database Systems. 7ed. Pearson Higher Education, Hoboken, NJ, EUA.

Ghotiya, S.; Mandal, J.; Kandasamy, S. 2017. Migration from relational to NoSQL database. IOP Publishing 263(4): 042055.

Heine, G. P.; Woltron, T.; Wöhrer, A. 2018. Towards a Scalable Data-Intensive Text Processing Architecture with Python and Cassandra. DATA ANALYTICS (7): 26-29.

Malik, A.; Burney, A.; Ahmed, F. 2020. A comparative study of unstructured data with SQL and NO-SQL database management systems. Journal of Computer and Communications, 8(4): 59-71.

Sadalage, P. J.; Fowler, M. 2013. Capítulo 1 – Por que NoSQL?. p. (25-39). In: Sadalage, P. J.; Fowler, M. NoSQL Essencial: Um Guia Conciso para o Mundo Emergente da Persistência Poliglota. 1ed. Novatec Editora, São Paulo, SP, Brasil.

Sethi, S.; & Panda, S. 2024. SQL or NoSQL – Practical Aspect and Rational behind Choosing Data Stores. Journal of Computer and Communications 12(8): 155-174.

Vilaça, R.; Cruz, F.; Pereira, J.; Oliveira, R. 2013. An effective scalable SQL engine for NoSQL databases. In: Held as Part of the 8th International Federated Conference on Distributed Computing Techniques, June 3-5, Florence, Italy. Anais… p. 155-168.

Article originating from the Final Course Work of the Specialization in Software Engineering of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform

You may also like

November 17, 2026

Detection and Diagnosis of Diseases in Corn and Rice in Mozambique with Deep Learning

Mozambican agriculture depends on corn and rice, fundamental crops for food security, but their productivity is strongly affected by phytosanitary diseases, whose early and accurate identification remains a challenge in resource-limited contexts. The study aimed to develop and compare Deep Learning-based models for the automatic detection and diagnosis of diseases in corn and rice leaves in Mozambique. Three approaches were used: a classic Convolutional Neural Network (CNN) developed from scratch, and two transfer learning models, ResNet50 and MobileNetV2. The dataset consisted of images from public databases and collected in the field, which underwent cleaning, balancing, and artificial augmentation steps. The models were trained and evaluated with standard performance metrics, including accuracy, precision, recall, F1-score, Log Loss, and Cohen’s Kappa. The classic CNN presented the best overall performance, with 94% precision, recall, and F1-score, and 93% accuracy and Cohen’s Kappa. MobileNetV2 achieved 91% F1-score and 90% accuracy, standing out for its lower Log Loss (0.30) and training time (6h32min59s). ResNet50 reached 80% F1-score and 79% accuracy. It was concluded that Deep Learning techniques have high potential to support phytopathological diagnosis in real agricultural contexts, contributing to faster and more accurate decisions in the management of corn and rice diseases.

Keywords: Deep learning; Rice; Plant diseases; Corn; Computer vision.

November 17, 2026

Artificial intelligence in accounting operational routines and financial efficiency

The incorporation of Artificial Intelligence (AI) and automation tools has reconfigured the routines of Brazilian accounting offices, impacting productivity, error mitigation, and financial efficiency. These impacts were analyzed in a small accounting firm, specialized in serving service providers in the medical field in São Caetano do Sul, São Paulo. The research adopted a qualitative approach, with a single case study design and incorporated units of analysis. Data were collected through a structured questionnaire, applied via Google Forms between January and February 2026, to five professionals in the operational structure of the firm in a census survey. The treatment followed content analysis with a priori and emergent categories. All respondents declared the use of automated tools or tools with AI resources. The perception of reduced execution time, decreased operational errors, rework, and better use of the team’s time converged in 80% of affirmative responses and 20% partial responses. The most benefited activities were bank reconciliation, import of fiscal documents, and tax assessment, with visible gains on typing errors, duplicate entries, and incorrect classification. Human intervention remained indispensable in tax planning, scenario analysis, and verification of ancillary obligations. The overall evaluation of AI’s contribution was predominantly favorable, with moderation linked to the complexity of fiscal regulations.

Keywords: Content analysis; Process automation; Accounting 4.0; Small business; Digital transformation.

November 17, 2026

Factors influencing schedule management in sewage treatment plant “retrofit” projects

Projects involving “retrofit” in Sewage Treatment Plants (STP) are characterized by high complexity, operational restrictions, and multiple technical interfaces. The study analyzed the perception of the professionals involved regarding the management and adherence to deadlines in “retrofit” projects for sewage treatment plants, considering technical, operational, and organizational factors. The research was characterized as an applied study, with a quali-quantitative approach, designed as a field survey. Data analysis was performed using descriptive statistics and qualitative categorization, employing SWOT, risk matrix (Probability x Impact), and 5W2H action plan tools. The results indicated a high perception of clarity (77.8%) and realism of deadlines (100% agreement), evidencing consistent initial planning. However, 77.7% of respondents pointed out that rework and scope changes directly impacted deadline adherence. The integration between engineering, manufacturing, and Production Planning and Control (PPC) showed an intermediate evaluation (77.7% partial/total agreement), indicating fragility in interfunctional coordination. The risk matrix highlighted as critical: scope changes, operational limitations of the STP, and technical complexity, all with a high level. As a central improvement, the standardization of scope change control, analysis, and approval was proposed, aiming to reduce rework and increase schedule predictability.

Keywords: Performance; Integration; Planning; Rework; Risk.

November 17, 2026

B2C Acquisition in diagnostic medicine: strategic partnerships with specialized medical offices

The diagnostic medicine sector in Brazil operates with high dependence on health plan providers, which limits the financial autonomy of service providers. This study proposed and analyzed a direct patient acquisition model (B2C) for a highly complex diagnostic medicine institution, structured through partnerships with Oncology and Cardiology practices and operationalized by a personalized customer service called concierge. The investigation adopted a mixed-methods approach in three phases: a field survey with a structured questionnaire applied to 87 patients, mapping of the operational flow using “Service Blueprint”, and analysis of the conversion rate from 312 medical referrals registered in the pilot project. The results indicated that trust in medical referral achieved 92.0% agreement, establishing itself as the main factor for adherence. The convenience of performing the exam on the same day reached 78.2%, and the willingness to pay directly was 55.2%, with polarization associated with the predominance of health plan users. The “Service Blueprint” identified three critical points in the transitions between stages of the patient journey. The overall conversion rate was 51.0%, with superior performance in Oncology (58.2%) compared to Cardiology (45.0%), and the greatest loss was concentrated in the acceptance and scheduling stage (28.3%). It was concluded that the model demonstrated possible operational viability and indications of patient acceptance, conditioned by price transparency and a reduction in the interval between consultation and approach by the concierge.

Keywords: Direct payment; Patient journey; Supplementary health; Concierge service; Conversion rate.

November 17, 2026

The role of the school manager in promoting an inclusive culture: leadership and educational policies

The research analyzed the school manager’s performance in promoting an inclusive culture, considering the need to transform legal guidelines and pedagogical foundations into effective institutional practices. It was assumed that school inclusion requires planning, organization, and strategic leadership in the school’s daily operations. A qualitative approach of an applied nature was adopted, developed in three complementary stages: documentary analysis of Brazilian legislation on inclusive education, bibliographic review on school leadership and management, and data systematization using management tools such as SWOT analysis and the PDCA cycle. The results indicated that the school management’s performance directly influenced the consolidation of inclusive environments, especially in the organization of processes, continuous team training, and strengthening of a collaborative institutional culture respectful of diversity. Structural, pedagogical, and relational challenges that demanded planned interventions were also identified. Finally, a Practical Guide for School Managers was developed, with guidelines, protocols, and operational orientations for decision-making and the organization of inclusive processes. It was concluded that the construction of an inclusive culture depended on the management’s ability to articulate public policies, pedagogical practices, and institutional processes, consolidating inclusion as a permanent commitment in the school context.

Keywords: educational accessibility; educational equity; participatory management; strategic planning; institutional processes.

November 17, 2026

An AI artifact based on Design Thinking to support the elicitation of software requirements from user needs

The development of software oriented to user needs frequently faces challenges in requirements elicitation, stemming from ambiguities in natural language and communication failures. Given this, an intelligent artifact integrating Design Thinking and Artificial Intelligence was proposed and validated, focusing on the Empathy and Definition stages, to support the collection, interpretation, and structuring of user needs. The research, of an applied, qualitative, and exploratory nature, conducted a Proof of Concept in a controlled environment, employing hypothetical scenarios. The artifact operated through guided natural language interviews, organizing information into elements such as pain points, context, and impact, and generating traceable problem syntheses. The results indicated that the system organized unstructured reports, reduced ambiguities, and preserved the meaning of the original information in most cases. The ability to reorganize logically in complex scenarios was also observed, although with limitations in separating multiple problems and dependence on instructions. It was concluded that the artifact has potential as a support for requirements elicitation, improving the clarity and structuring of information, but requires supervision and contextual control.

Keywords: Design Thinking; Requirements Elicitation; Requirements Engineering; Artificial Intelligence; Natural Language Processing.

November 17, 2026

The Extinction of the Monophasic Tax System in the Pharmaceutical Industry in Light of the Tax Reform

The consumption Tax Reform, established by Constitutional Amendment No. 132/2023, introduced structural changes to the Brazilian tax system, with the replacement of the Social Integration Program (PIS) and the Social Security Financing Contribution (COFINS) by the Contribution on Goods and Services (CBS), in a dual Value Added Tax (VAT) model. The objective was to analyze the impacts of the extinction of the single-stage system on the pharmaceutical industry, in light of the new regulatory design and its economic and legal effects. A qualitative, exploratory research was conducted through a case study and documentary analysis of legislation and specialized doctrine. It was examined whether the end of the single-stage system would imply a relevant increase in the tax burden for the sector. The results indicated that, although the formal overcoming of the single-stage system has occurred, the material exemption of medicines has been preserved, mainly due to the application of reduced rates of 60% and, in certain cases, a zero rate of the Tax on Goods and Services (IBS) and CBS. The change in model may cause occasional variations in revenue collection, without characterizing a structural increase in the tax burden, tending towards neutrality over time. It was preliminarily concluded that the Tax Reform reorganized the taxation technique in the pharmaceutical industry, replacing a concentrated regime with more transparent exemption mechanisms. This approach proved compatible with the principles of neutrality, full non-cumulativeness, and fiscality, maintaining tax protection for essential medicines and aligning fiscal policy with the fundamental right to health.

Keywords: CBS; COFINS; IBS; Dual VAT; PIS.