Vercel Unveils Fluid Computing to Unify All Workloads

Vercel has unveiled Fluid Computing, a system that unifies diverse workloads such as builds, sandboxes, and functions. Previously, different computing infrastructures were required for each type of task, but Fluid assembles machines according to workload characteristics, dynamically changes configurations, and absorbs burst capacity in real time.

Scale and Composition of Fluid

Fluid currently handles 15 million builds per day, 25 million sandboxes per week, and 1 trillion requests per month. The system consists of the execution model, Fluid Computing, along with hardware (Hive), environments (Fluid images), and storage (Vercel Drives).

Execution by Workload

Fluid instantly provides optimized machines based on the needs of the workload.

• Builds: Compute-bound tasks running on high-spec machines with abundant CPU and memory
• Functions: IO-bound tasks running on small VMs that load code instantly upon request
• Sandboxes: Tasks with flexible configurations connected to Drives containing user data

Each workload runs in isolated VMs provided by Hive (mostly pre-warmed), and storage is decoupled from the machines to ensure data accessibility. Fluid Computing improves efficiency by handling multiple requests on a single instance and applying Active CPU billing.