Analisis Performa Kubernetes K3S dan K8S pada Pengolahan Data Citra Medis untuk Optimalisasi Sumber Daya

Hannisa Ada Fitria, Dr. Ir. Ronald Adrian, S.T., M.Eng., IPM.

2026 | Tugas Akhir | D4 TEKNOLOGI JARINGAN

Perkembangan kontainerisasi menjadikan Kubernetes platform dominan, namun arsitektur standar (K8S) kurang efisien pada perangkat keras terbatas. K3S hadir sebagai alternatif ringan untuk mengakomodasi lingkungan edge computing. Penelitian ini membandingkan kinerja K3S (native) dan K8S (KinD) dalam mengorkestrasi layanan pengolahan citra medis menggunakan dataset X-Ray Pneumonia dengan tiga variasi rasio split dataset (50:50, 60:40, 70:30) pada Virtual Private Server (VPS). Metrik yang dievaluasi mencakup latency, Requests per Second (RPS), serta utilisasi CPU dan memori. Layanan ini dibangun menggunakan kerangka kerja Flask, OpenCV, dan NumPy di dalam lingkungan Virtual Private Server (VPS). Simulasi pengujian beban jaringan dilakukan menggunakan Apache Benchmark (10-500 request) membuktikan K3S lebih unggul dengan rata-rata 43,55 RPS dan latensi 401,59 ms, jauh melampaui K8S yang tertahan pada 16,23 RPS dan latensi 1.130,37 ms. K3S mampu memaksimalkan CPU hingga 59,52% saat beban puncak, sementara K8S mengalami bottleneck meskipun lebih hemat penggunaan RAM. Dapat disimpulkan bahwa K3S dengan rasio model 50:50 merupakan solusi paling optimal untuk layanan medis pada infrastruktur menengah dengan prioritas kecepatan, sedangkan K8S tetap relevan untuk skenario enterprise yang memiliki restriksi memori ketat.

 

The rapid development of containerization has established Kubernetes as the dominant platform; however, its standard architecture (K8S) is often inefficient on limited hardware resources. K3S emerged as a lightweight alternative designed to accommodate edge computing environments. This study compares the performance of native K3S and K8S (KinD) in orchestrating medical image processing services using an X-Ray Pneumonia dataset with three train-test split ratio variations (50:50, 60:40, and 70:30). The service was developed using Flask, OpenCV, and NumPy frameworks within a Virtual Private Server (VPS) environment. The evaluated metrics include latency, Requests per Second (RPS), as well as CPU and memory utilization. Network load testing simulations conducted using Apache Benchmark (ranging from 10 to 500 requests) demonstrated that K3S significantly outperformed K8S. K3S achieved an average throughput of 43.55 RPS with a latency of 401.59 ms, far surpassing K8S, which plateaued at 16.23 RPS with a latency of 1,130.37 ms. Furthermore, K3S was able to maximize CPU utilization up to 59.52% during peak loads, whereas K8S experienced a bottleneck despite being more efficient in RAM consumption. In conclusion, K3S combined with a 50:50 model ratio provides the most optimal solution for medical services on mid-tier infrastructure where processing speed is prioritized. Conversely, K8S remains relevant for enterprise-scale scenarios constrained by strict memory limitations.

Kata Kunci : Evaluasi Kinerja, K3S, Kubernetes, Orkestrasi Kontainer, Virtual Private Server.