Description
The recent rise in popularity of serverless computing has brought forth new challenges in determining the right consistency model for applications. Previous work has studied how to bring scalable, transactional isolation to serverless computing, but has been insufficient at preventing several consistency anomalies. To address this problem, we would like to guarantee stronger transactional isolation for serverless computing in a scalable manner.
In this thesis we present TASC, a transactional shim for serverless applications. TASC offers flexibility in interposing between most commodity FaaS platforms and cloud storage engines. It providers stronger transactional isolation by guaranteeing snapshot isolation. TASC decouples transaction management from data management and uses modified read and commit protocols to provide scalable snapshot isolation. We demonstrate that TASC has comparable overhead to other serverless consistency shims, and manages to scale to thousands of transactions per second.