Struggling with slow or unreliable integration tests? This case study shows how one company reduced test execution time from hours to minutes while improving stability across thousands of database combinations.
Context: Real-Time Data Integration in a Complex Ecosystem
MOLO17 is a B2B software company founded in 2016, specializing in the development of real-time data integration solutions for complex enterprise environments.
With operational offices in Italy and the United States and an internationally distributed team, MOLO17 supports companies managing heterogeneous data environments (data from different systems and formats), helping them overcome the limitations of legacy systems and data silos.
At the time, many enterprise organizations were still operating within data silos based on relational databases and legacy systems (RDBMS, mainframe, AS/400), which were often difficult to integrate and tightly controlled by the teams responsible for their maintenance.
At the same time, the market was adopting more modern and scalable architectures, such as NoSQL databases, creating a significant gap between traditional systems and new data platforms.
To bridge this gap, MOLO17 developed Gluesync, a data replication and integration platform that enables real-time data transfer from legacy systems to NoSQL databases and, over time, to an increasing number of modern platforms:
- RDBMS ↔ RDBMS
- RDBMS ↔ NoSQL
- Legacy systems ↔ Data Lake
- Event streaming to platforms such as Snowflake, Azure Data Lake, and Databricks
Today, Gluesync supports more than 35 different databases and is used globally in enterprise and mission-critical environments.
Why Integration Tests Become Slow at Scale
As integration environments grow, test execution often slows due to I/O bottlenecks, limited storage performance, and the complexity of running multiple database combinations in parallel.
In large-scale systems, integration testing is not only a question of CPU capacity. Disk performance, data consistency, and the ability to handle parallel workloads play a critical role. When these factors are not aligned, test execution becomes slower, less predictable, and harder to scale.
This is especially true in environments where multiple databases, versions, and configurations must be tested continuously.
The Challenge: Reliably Testing Thousands of Combinations
As the platform grew, a clear technical need emerged: continuously testing a vast number of database–version–configuration combinations.
Over the past few years, MOLO17 developed a suite of automated integration tests to ensure that every new feature released was compatible with all supported combinations. The workload is extremely intensive.
Integration test performance is often limited by disk I/O rather than CPU capacity. This becomes especially clear in workloads such as:
- Dozens of databases started in parallel
- Massive disk write operations
- Repeated and orchestrated queries
- Thousands of combinations tested for each build
Before the new cluster, the team operated on a 6-node infrastructure with around 300 CPU cores, but with significant storage and I/O limitations:
- Insufficient IOPS
- Disk instability
- Unpredictable and non-repeatable test runs
- Frequent manual restarts
- Test durations ranging from 1 to 2 hours
The core issue was not just performance, but reliability. When tests fail due to infrastructure limitations, results become unusable and the entire process loses its value.
Common Causes of Unreliable Test Results
Unreliable test results are often caused by infrastructure limitations rather than issues in the test logic itself. When environments rely on insufficient disk performance, unstable storage, or limited parallel processing, test outcomes become inconsistent and difficult to reproduce.
In high-scale integration testing, disk I/O becomes a critical bottleneck. Running multiple databases in parallel generates intensive read and write operations, which can quickly exceed the capabilities of standard infrastructure.
To address these issues, organizations need to focus on improving storage performance, increasing parallel execution capacity, and ensuring stable environments that can be consistently deployed and reset.
How to Improve Integration Test Performance at Scale
Improving integration test performance requires a combination of high I/O infrastructure, properly configured server environments, parallel execution capabilities, and stable environments that can be quickly deployed and reset.
Without these elements in place, even well-designed test suites can become slow, unreliable, and difficult to scale in production environments.
The Technical Goal: Speed, Stability, and Trust in Testing
MOLO17 needed an infrastructure capable of:
- Supporting extremely high I/O workloads
- Ensuring test stability and repeatability
- Quickly spinning up and tearing down database environments
- Drastically reducing execution times
- Integrating hardware already purchased
- Maintaining an on-prem architecture with a cloud-like experience
All of this without exceeding budget constraints, at a time when component costs (NAND, storage) were rising.
Why Renewtech: Technical Approach and Tailor-Made Design
Selecting a hardware partner was not immediate. The evaluation process lasted about three months and was based on a technical scorecard that included:
- Total cost
- Technical support
- Ability to reuse existing components
- Engineering quality
- Response times
- Delivery times
Renewtech stood out not only for staying within budget, but for its ability to propose concrete technical solutions, adapting available hardware to the real needs of the workload.
A key element was integrating disks already purchased by MOLO17 into the new architecture, avoiding waste and protecting previous investments.
The delivery of the final critical node took just two days, allowing the team to complete the cluster quickly and go live without further delays.
The Result: 16,000 Tests in 10–12 Minutes
Once in production, the new cluster radically transformed the engineering team’s workflow.
Performance Comparison:
| Factor | Before | After |
| Test duration | 1-2 hours |
10-12 minutes |
| Stability | Unpredictable | Reliable |
| Parallel databases | Limited | 80-90 |
Today, MOLO17 can:
- Start 80–90 databases in parallel
- Run more than 16,000 integration test combinations
- Complete the full test cycle in 10–12 minutes
- Achieve stable and reliable results
The key shift was from unpredictability to trust in testing.
This enables the team to:
- Validate a new build immediately
- Iterate multiple times per day
- Release features continuously
- Avoid risky overnight deployments across time zones
In practice, the infrastructure made MOLO17’s target development model sustainable: one new feature per day, with reliability.
On-Prem, with a Modern Mindset
A distinctive aspect of the solution is that the entire platform operates on-prem, while offering a cloud-like experience:
- Over-the-air updates
- Direct control over machine power and usage
- Autonomous workload and energy management
This approach allows MOLO17 to maintain full control over its environment, reducing operational complexity and external dependencies.
Technical Collaboration and Shared Vision
From MOLO17’s perspective, Renewtech’s value went beyond hardware supply. It lay in having access to a skilled engineering team capable of:
- Understanding complex workloads
- Guiding technical decisions
- Supporting organizations of different sizes
- Maintaining response and delivery times aligned with enterprise standards
An approach fully consistent with what MOLO17 itself delivers to its customers.
Conclusion
This case shows that, in complex software environments, infrastructure is not a mere operational detail, but a key enabler of the development model.
When testing, hardware, and processes are aligned, the result is not just speed.
It is reliability.
FAQ: Frequently Asked Questions
The following questions often come up when dealing with large-scale integration testing environments:
Why are integration tests slow?
Integration tests are often slow due to limited disk I/O, insufficient parallel processing, and unstable environments.
How can integration test performance be improved?
By using high-performance storage, running tests in parallel, and ensuring stable infrastructure.
Is on-prem infrastructure better for testing?
In high-I/O workloads, on-prem infrastructure can provide more predictable performance and lower long-term costs compared to cloud environments.