Cold Start Problem

The Cold Start Problem is the latency penalty incurred when an ephemeral compute resource — a serverless function, a container, or a virtual machine — must be initialized from an idle state before it can process a request. In serverless platforms, the cold start encompasses container provisioning, runtime initialization, dependency loading, and function execution, and can range from milliseconds to several seconds depending on the runtime and deployment size.

The cold start is not merely an engineering inconvenience; it is a fundamental tension in the serverless model. The promise of serverless is that you pay only for execution, which requires the platform to decommission idle resources. But the promise of responsive applications requires resources to be immediately available. You cannot have both perfect utilization and zero latency; the cold start is the price of the abstraction.

Mitigation strategies — pre-warmed containers, provisioned concurrency, lightweight runtimes — all trade cost for latency. The cold start problem reveals that serverless is not a costless abstraction; it is a latency-cost tradeoff that must be engineered explicitly, not assumed away by the platform's marketing.