TITLE:
A Comparative Analysis of Static Scaling, Kubernetes Horizontal Pod Autoscaling (HPA), and Kubernetes Event-Driven Autoscaling (KEDA) in Moodle LMS
AUTHORS:
Robert Achang’a, Elisha Abade, Andrew Kahonge
KEYWORDS:
Kubernetes, Horizontal Pod Autoscaler, Kubernetes Event-Driven Autoscaler, Moodle, Autoscaling, Online Examination, Performance Evaluation, Peak Workload
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.10,
October
21,
2026
ABSTRACT: Moodle is widely used to administer online examinations, but fixed-capacity deployments can become unstable when many learners log in, open quiz pages, and submit responses within the same examination window. This study evaluates three deployment strategies for a Moodle-based examination platform running on an on-premises Kubernetes cluster: static scaling, Kubernetes Horizontal Pod Autoscaling (HPA), and Kubernetes Event-Driven Autoscaling (KEDA). Apache JMeter generated workloads from 500 to 70,000 concurrent connections. Performance was assessed using average response time, 95th percentile latency, 99th percentile latency, throughput, and error rate. Static scaling degraded earliest: average response time increased from 328 ms at 500 concurrent connections to 10,347 ms at 2500 concurrent connections, while the error rate rose from 0% to 81% over the same interval. HPA produced the strongest reliability profile, maintaining 0% errors from 500 to 5000 concurrent connections and recording the lowest error rate among the three strategies at every workload from 10,000 to 70,000 concurrent connections. KEDA produced the strongest responsiveness among successful requests at several higher-load points, including 412 ms average response time at 10,000 concurrent connections and 1604 ms at 70,000 concurrent connections, and it achieved the highest throughput at most large workloads. However, KEDA’s error rate increased sharply after 5,000 concurrent connections, reaching 93.725% at 10,000 concurrent connections. The findings show that autoscaling improves Moodle performance compared with static provisioning, but HPA and KEDA optimize different parts of the performance envelope. For high-stakes online examinations, HPA is the safer reliability-oriented option in this dataset, while KEDA is promising when responsiveness and throughput are prioritized and its triggers are carefully tuned.