EVALUATION OF SERVERLESS COLD START OPTIMIZATION TECHNIQUES

Abstract

Cold start latency remains one of the main challenges in serverless computing, particularly for Function-as-a-Service (FaaS) applications operating under dynamic workloads. Runtime initialization, dependency loading, and resource allocation increase response latency and reduce system performance. Although numerous cold start mitigation techniques have been proposed, comprehensive experimental comparisons under identical workload conditions remain limited. This study analyzes contemporary cold start optimization techniques and experimentally evaluates the effectiveness of warm-up, caching, and concurrency strategies for improving serverless application performance under different workload intensities. A controlled serverless test environment was developed using Python, Flask, Redis, and Locust. Cold starts were simulated through artificial initialization delays, while Redis was used for result caching and a thread pool enabled concurrent request processing. Performance was evaluated under moderate (100 users), intensive (300 users), and prolonged high-load scenarios using response time, throughput, request-processing errors, and P95/ P99 latency metrics. The results show that optimization effectiveness depends on workload intensity. Under moderate workloads, performance differences were minor. Under intensive workloads, warm-up and concurrency reduced response latency and improved system stability. Caching reduced the error rate in the short-term intensive scenario, although it introduced substantial additional latency. During prolonged testing, the average response time of the caching configuration decreased, indicating a more workload-dependent effect of cache population. Combined optimization provided balanced performance but did not consistently outperform individual techniques because of coordination overhead. The findings provide practical guidance for selecting cold start optimization strategies according to workload characteristics and support the design of adaptive, high-performance serverless applications and cloud services.