State-of-the-Art Advancements in Automated Storage Capacity Forecasting and Time-Series Modeling for Large-Scale Relational Engines
Journal of Contemporary Science and Technology Management · 16.12.2025 · 1 (04) · 592-636
DOI: 10.66592/2vhmah74 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Contemporary Science and Technology Management, online erschienen am 16.12.2025. Verfasst von Evans Ighodalo und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Running out of storage remains one of the most common and most avoidable causes of outages in large relational database estates. For decades, capacity planning for engines such as Oracle Database, Microsoft SQL Server, PostgreSQL and MySQL relied on linear extrapolation of tablespace or file growth, reviewed manually by database administrators. The year 2025 marked a turning point. Pretrained time-series foundation models (TSFMs) such as TimesFM 2.5, Chronos-2, Moirai 2.0 and Toto matured into compact, fast and openly available forecasters capable of zero-shot prediction on telemetry they were never trained on. In parallel, distribution-free uncertainty quantification through conformal prediction was adapted to non-exchangeable, multi-step time series, and research on self-driving and agentic database management began coupling forecasts directly to automated actions. Using an integrative literature review of 254 sources published up to 2025, a conceptual model and an illustrative simulation, this paper examines these developments as they apply to automated storage capacity forecasting for large-scale relational engines. We characterise the growth dynamics that make database storage series unusual (structural breaks from tenant onboarding, step changes from retention purges and bulk loads, MVCC bloat, and log churn), review classical, machine learning and foundation-model approaches, discuss the 2025 benchmarking debate over data contamination, and propose a six-layer conceptual model that separates forecasting, calibration and decision layers. An illustrative experiment on 300 synthetic database growth series shows that recency-aware and change-point-aware models reduce 90-day sMAPE from 6.14% to 1.17% relative to full-history linear regression and raise exhaustion-alert recall from 19.6% to 80.4%, while also showing that naive Gaussian intervals badly under-cover and that even horizon-wise conformal intervals fall short of nominal coverage under drift. We also situate these advances within a broad cross-domain literature on predictive analytics, anomaly detection, IT operations, security and AI governance comprising more than 200 related studies. We close with open problems and practical guidance.
Zitieren
APA
Ighodalo, E., Kolawole, A., & Ogundapo, O. (2025). State-of-the-Art Advancements in Automated Storage Capacity Forecasting and Time-Series Modeling for Large-Scale Relational Engines. Journal of Contemporary Science and Technology Management, 1(04), 592-636. https://doi.org/10.66592/2vhmah74
Vancouver
Ighodalo E, Kolawole A, Ogundapo O. State-of-the-Art Advancements in Automated Storage Capacity Forecasting and Time-Series Modeling for Large-Scale Relational Engines. Journal of Contemporary Science and Technology Management. 2025;1(04):592-636. doi:10.66592/2vhmah74
BibTeX
@article{Ighodalo2025,
author = {Ighodalo, Evans and Kolawole, Ayinoluwa and Ogundapo, Oluranti},
title = {State-of-the-Art Advancements in Automated Storage Capacity Forecasting and Time-Series Modeling for Large-Scale Relational Engines},
journal = {Journal of Contemporary Science and Technology Management},
year = {2025},
volume = {1},
number = {04},
pages = {592--636},
doi = {10.66592/2vhmah74}
}
EUVIMED