Alto Snapshot/Restore: Enabling Stateful, Long-Lived Serverless Computation

As demand for the serverless computing paradigm increases, so does demand for a currently unavailable serverless architecture: stateful, long-lived serverless computation. In this paper, we re-introduce Alto, which provides such an architecture. Using Alto, developers can enjoy the autoscaling and fine-grained billing promised by the serverless platform while also enjoying the performance benefits of co-locating code and local data. This serverless paradigm makes developing existing applications easier and also enables new cloud application architectures. We explain why providing stateful, long-lived serverless computation requires fast snapshot and restore of computation. We explain how Alto virtualizes the managed runtime to provide fast snapshot and restore. We describe how Alto moves filesystem and network state from the kernel into the Alto runtime to improve performance. Finally, we evaluate an Alto prototype against Checkpoint-and-Restore In Userspace (CRIU), a fast process checkpoint-restore tool, and show that Alto snapshot and restore performance is comparable.