külmstart

Külmstart, in Estonian often translated as "cold start," is a term used to describe the initial startup of a system when there is little or no prior state, data, or cached information.

In information technology, the cold start problem is common in recommender systems, search and indexing, and machine-learning models. It manifests in several forms, including new-user cold start (no user history), new-item cold start (no interactions with a new item), and system-wide cold start (no cached results or trained models). It contrasts with warm start, where some context or data exists, and hot start, where updates arrive rapidly from recent activity. Addressing cold start typically involves strategies to gather initial data, leverage content-based signals, apply transfer learning, or use hybrid approaches that combine multiple sources of information.

In cloud and serverless computing, a cold start refers to the latency that occurs when a function or service is invoked after being idle. The runtime environment must be provisioned, dependencies loaded, and code initialized, leading to an initial delay before the function can respond. Mitigation includes keeping a small number of ready-to-use instances, deploying lightweight initializations, or scheduling activations to avoid long idle periods.

In automotive contexts, a cold start describes starting an internal combustion engine in cold ambient temperatures. Engine oil viscosity, fuel-air mixture, and battery performance can influence starting reliability, engine wear, and emissions. Practical considerations include using appropriate engine oil, maintaining battery health, and allowing a brief warm-up period in very cold conditions.

Across domains, common mitigation approaches involve improving initial data collection, employing hybrid methods, pre-warming systems, and optimizing initialization workflows to reduce start-up time and improve early performance.