Kernel-Level Performance Evaluation of Windows Server 2022 During High-Concurrency Access to Educational Applications in the Schools Division Office of Passi City

Institutional web application infrastructure within regional educational administrative sectors frequently relies on monolithic web runtimes deployed natively on single bare-metal compute hosts. This study presents an empirical, kernel-level performance evaluation of the localized bare-metal server infrastructure orchestrating daily logistics for the Schools Division Office (SDO) of Passi City. The environment hosts a production suite of enterprise educational and administrative platforms—including DocWatch (Document Tracking System), Baol sang Kaalam (Localized Educational Resources), and BioSync (Attendance Management System)—operating natively on an Apache, PHP, and MariaDB (XAMPP) stack. To bypass impractical hypervisor abstraction layers, this research applies an experimental research design to evaluate core operating system kernel mechanics under stress. Specifically, it measures symmetric multiprocessing (SMP) process management, thread scheduling frequencies, and memory working set boundaries under high-concurrency conditions. Using Apache JMeter (Apache Software Foundation, 2024), the ecosystem was subjected to parameterized concurrent thread loads ranging from 100 to 1,000 users. An initial baseline test revealed that reaching 1,000 users triggered severe thread pool exhaustion, causing transaction error rates to rise to 3.47% despite average CPU saturation dropping to 40.70%. To resolve this operational blockage, a subsequent multi-iteration ablation study was executed by applying targeted Multi-Processing Module (MPM) and database buffer optimizations. These targeted interventions successfully resolved the hardware-level bottleneck, reducing process lock contention and allowing the operating system to comfortably sustain 1,000 concurrent users with a near-zero transaction error rate (<0.01%) while stabilizing average CPU utilization down to 31.66%. Building upon these successful optimization outcomes, this study proposes a Dynamic Thread-Scaling Framework (DTSF)—a zero-cost, software-defined automation framework designed in alignment with modern ISO/IEC performance efficiency standards (International Organization for Standardization, 2024) to dynamically adjust thread allocation vectors based on real-time OS performance counters. Finally, an unmitigated boundary stress test utilizing a synthetic extreme load of 10,000 concurrent users successfully identified the absolute hardware exhaustion limit of the physical compute node, characterized by sustained 100% CPU utilization and a 17.56% application failure rate. The captured empirical telemetry serves as a data-driven configuration blueprint proving that low-level kernel optimization can significantly extend physical infrastructure lifecycles in resource-constrained public-sector ecosystems.