Faabric: Shared Memory and Message Passing for Serverless

Serverless providers make fine-grained changes to an application’s parallelism and distribution at runtime to save users time and money. This changes come at a high cost: serverless applications today must be written using stateless ephemeral functions. Stateless emphemeral functions can not efficiently share data, nor communicate between each other, ruling out the adoption of two enduringly popular programming models: shared memory and message passing. We present Faabric, a serverless runtime that supports shared memory and message passing, while retaining control of parallelism and distribution.