Function as a Service
Function as a Service (FaaS) is a cloud computing model in which developers deploy individual, event-triggered functions rather than whole applications or servers, and the platform automatically handles scaling, execution, and billing per…
Definition
Function as a Service (FaaS) is a cloud computing model in which developers deploy individual, event-triggered functions rather than whole applications or servers, and the platform automatically handles scaling, execution, and billing per invocation. It is the most common concrete implementation of the broader serverless computing model.
Overview
FaaS platforms — most notably AWS Lambda (which popularized the category starting in 2014), Google Cloud Functions, Azure Functions, and Cloudflare Workers — let developers write and deploy a single function, package its dependencies, and configure it to run in response to a specific trigger: an HTTP request via an API gateway, a new object uploaded to cloud storage, a message arriving on a queue, a scheduled cron-style timer, or a database change stream. The platform handles everything needed to actually execute that function at scale: provisioning execution environments on demand, running potentially thousands of concurrent invocations in parallel without the developer managing any of that concurrency infrastructure, and tearing environments back down when no longer needed. Billing in FaaS is typically metered extremely granularly — by the millisecond or hundred-millisecond of actual execution time multiplied by allocated memory, plus a small per-invocation fee — meaning a function that runs rarely costs close to nothing, while one invoked millions of times per day scales its cost roughly linearly with usage. This pay-per-use model, combined with zero idle cost, is the primary economic argument for FaaS over maintaining an always-on server for the same workload, particularly for unpredictable or bursty traffic patterns. FaaS functions are intentionally constrained: most platforms impose maximum execution duration limits (historically 15 minutes for AWS Lambda, for example), limited or ephemeral local disk/filesystem access, and a 'stateless by design' expectation, since any given invocation might run in a fresh environment with no memory of prior invocations, requiring external storage (a database, cache, or object store) for any state that needs to persist. Cold starts — the latency incurred spinning up a fresh execution environment for a function that hasn't run recently — remain one of FaaS's most discussed limitations, particularly for latency-sensitive applications, though providers have invested heavily in reducing cold-start times through techniques like provisioned concurrency (pre-warmed instances kept ready) and lighter-weight runtimes.
Key Features
- Deploys individual event-triggered functions rather than whole applications
- Automatic scaling to handle concurrent invocations without developer-managed infrastructure
- Billing metered by execution time and memory, typically at sub-second granularity
- Triggered by HTTP requests, storage events, queue messages, or scheduled timers
- Stateless execution model — persistent state must live in external storage
- Enforced maximum execution duration limits per invocation
- Cold-start latency when spinning up a fresh execution environment
- Provisioned concurrency and lighter runtimes used to mitigate cold starts