Reducing cold-start delays in FaaS and serverless environmentsPodobnik, Teodor Janez (Avtor) Ciglarič, Mojca (Mentor) Pančur, Matjaž (Komentor) NetworkingContainerContainer orchestrationServerlessFaaScold start delayCloud ComputingBPFXDPOpenFaaSServerless computing has gained popularity in the cloud computing model by abstracting infrastructure management. Serverless functions (FaaS) further abstract backend infrastructure, allowing developers to focus solely on function implementation and code execution. Both serverless computing and FaaS rely on container primitives, packaging code and dependencies for flexible deployment. However, "cold starts" introduce significant latency when containers or functions are spawned for the first time. Existing research primarily focuses on container runtime bootstrap and probability based function prewarming, rather than its startup through the existing connection establishment. This thesis proposes appending a unique function identifier in TCP SYN packet header to pre-warm the functions container, using an eBPF library for secure and efficient program execution. The approach is integrated into open-source framework for evaluation and implementation in a real-world use case. The thesis aims to remove cold start delays entirely, thus aligning the response time with warm starts. This synchronization occurs when network latency matches the function's cold-start time in typical scenarios, a common occurrence in cloud environments and heavily trafficked IoT networks.20232023-09-29 13:35:00Magistrsko delo/naloga151108VisID: 35406COBISS_ID: 169474051sl