Azure Load Testing for Vaults in Cloud Based Environments
Irshad, Hamza (2026)
2026
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2026-06-16
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202606157503
https://urn.fi/URN:NBN:fi:tuni-202606157503
Tiivistelmä
This thesis examines how a repeatable and realistic performance testing framework can be built for a metadata driven document management platform whose behaviour depends on vault structure, workflows, permissions, storage configuration, and cloud deployment context. The work was carried out as an industrial case study at M-Files, a software company that develops such a platform. The practical problem is that ad hoc or sample based load tests do not provide enough evidence for comparing vault variants after infrastructure and storage changes, and that failed test runs still require costly manual diagnosis.
The study follows a constructive industrial case study design. A reusable artifact was designed and implemented around five connected parts: a realistic reference vault, an Application Programming Interface (API) based population tool, a repeatable execution toolchain built on Apache JMeter, Azure Load Testing, TeamCity, and Groovy, a comparative multi vault evaluation setup, and an evidence driven Artificial Intelligence (AI) failure analysis component. The empirical material consists of public product and tool documentation, scholarly literature, the author’s direct observations from the case work, and Azure Load Testing run summaries with their associated server side monitoring data.
The thesis shows that benchmark realism matters in vault based systems because metadata card complexity, workflows, permissions, and evolving content change what is actually measured under load. The unified toolchain makes comparison repeatable by reusing the same logical workload across vault variants through external parameters and version controlled execution. The empirical findings support the working hypothesis with qualifications: the realistic benchmark and unified execution chain improved comparability and diagnosis, while cloud-native configuration alone did not remove bottlenecks. In the complex reference vault run, database pressure remained the dominant bottleneck under the high load profile. As a result, in the master storage enabled run, latency was markedly lower, and the error rate dropped substantially.The AI failure advisor was shown to be practically integrable as a decision-support capability for categorising failures, suggesting likely causes, and presenting actionable next steps. The findings are case based and not statistically generalisable beyond the studied environment.
The main contribution is not one isolated load test but a reusable, case tested framework for realistic, comparative, and diagnostically useful performance testing in a vault based enterprise environment.
The study follows a constructive industrial case study design. A reusable artifact was designed and implemented around five connected parts: a realistic reference vault, an Application Programming Interface (API) based population tool, a repeatable execution toolchain built on Apache JMeter, Azure Load Testing, TeamCity, and Groovy, a comparative multi vault evaluation setup, and an evidence driven Artificial Intelligence (AI) failure analysis component. The empirical material consists of public product and tool documentation, scholarly literature, the author’s direct observations from the case work, and Azure Load Testing run summaries with their associated server side monitoring data.
The thesis shows that benchmark realism matters in vault based systems because metadata card complexity, workflows, permissions, and evolving content change what is actually measured under load. The unified toolchain makes comparison repeatable by reusing the same logical workload across vault variants through external parameters and version controlled execution. The empirical findings support the working hypothesis with qualifications: the realistic benchmark and unified execution chain improved comparability and diagnosis, while cloud-native configuration alone did not remove bottlenecks. In the complex reference vault run, database pressure remained the dominant bottleneck under the high load profile. As a result, in the master storage enabled run, latency was markedly lower, and the error rate dropped substantially.The AI failure advisor was shown to be practically integrable as a decision-support capability for categorising failures, suggesting likely causes, and presenting actionable next steps. The findings are case based and not statistically generalisable beyond the studied environment.
The main contribution is not one isolated load test but a reusable, case tested framework for realistic, comparative, and diagnostically useful performance testing in a vault based enterprise environment.