Analisis Performa Eviction Policy Redis Menggunakan Load Testing pada Pola Kueri Kuis

Abstract

Quizzes in the Information Technology program are often held simultaneously or at different times for each class, resulting in an increased number of requests to the database. The issue lies in the storage of quizzes using disk, necessitating the use of Redis to expedite the data retrieval process. The quiz query pattern involves a time-related variable, requiring Redis to effectively manage memory storage. To test the data eviction policies, the study employed load testing methods on Least Recently Used (LRU), Least Frequently Used (LFU), and Volatile-ttl. The researchers constructed a backend server to connect to SQL and Redis, providing specific responses indicating the source of the data to determine the cache hit ratio. An API was developed and subjected to load testing five times. Measurements were taken in terms of milliseconds per request and data retrieval time. The load test was conducted over a 50-minute period, gradually increasing concurrency levels up to 100 users to assess performance under different load conditions. The results of the study revealed that Volatile-ttl exhibited a significant time difference, being 27.2 seconds faster than LRU and LFU in terms of milliseconds per request and 215 ms faster in terms of data processing. Volatile-ttl also achieved the highest hit ratio at 88.42%, followed by LFU at 86.33% and LRU at 85.80%. Volatile-ttl demonstrated a 2.62% higher hit ratio compared to LRU and LFU, indicating that on average, Volatile-ttl processed requests with a cache of 208.496 out of 233.388 requests.