ASTRA : Advanced Storage Tiering, Resilience, and Analytics Framework

Abstract

Industrial and manufacturing computing environments rely heavily on distributed virtualized enterprise topologies running critical database systems, enterprise resource planning (ERP), and manufacturing execution pipelines. A critical point of failure in these deployments is the resilience and availability of the underlying backup repository ecosystem. Ad-hoc administrative storage management and manual capacity inspections create systemic risks, including silent storage exhaustion, unmonitored backup failure cascades, and unmitigated capacity bottlenecks. This paper presents the architecture, theoretical foundation, mathematical modeling, and production deployment of an autonomous storage analytics, multi-tier capacity forecasting, and notification platform. Engineered across geographically isolated plant nodes (A, B, C, D, and E), the proposed system leverages agentless, low-dependency CIM/WMI hardware polling coupled with recursive filesystem discovery. The platform integrates a rolling-window statistical ledger, automated schema persistence, and structured HTML/Excel dispatch protocols over encrypted transport layer security (TLS). Furthermore, to counter the degradation of standard linear regression caused by non-organic data operations (such as manual inter-tier storage migrations and retention overrides), we formulate an aggregate pooling abstraction and a rolling median smoothing mechanism. The empirical deployment demonstrates low runtime compute overhead (1.15% peak CPU, 26.4 MB memory) and substantially reduces manual, per-server capacity checks across all five plants. In a representative validation case involving an unplanned 420 GB inter-tier migration, the proposed macro-pooling and median-filtering approach reduced the seven-day capacity-forecast RMSE from 482.4 GB (standard linear regression) to 4.12 GB, illustrating the model's resilience to administrative storage-migration noise.