awesome-serverless-research

The paper presents AFT, an atomic fault tolerance shim for serverless functions. AFT ensures atomic visibility of updates by enforcing read atomic isolation guarantee.

The paper presents Beldi, a library and runtime system for writing fault-tolerant stateful serverless functions. Code available at

It is the best paper of OOPSLA'19, which presents formal model to analyze (and prove) the reliability of serverless computing.

This work introduces one system and one optimization. The system is vHive, an open-source and full-stack serverless framework. The framework is built based on several state-of-the-art infastructure, e.g., Firecracker. Besides, the paper proposes an optimization, called REAP, which is a software mechanism for serverless hosts that records functions' stable working set of guest memory pages and proactively prefetches it from disk into memory.

This paper presents ServerlessBench, which is the first and state-of-the-art serverless benchmarks. The benchmark suite is open-sourced at

SIW is an industry paper by Microsoft Azure. You can learn about real-world issues/techniques of serverless computing from it.

A curated list of awesome services, solutions and resources for serverless / nobackend applications.

A curated list of awesome serverless security resources such as (e)books, articles, whitepapers, blogs and research papers.

Boki solves many problems on stateful serverless computing, including state consistency, fault tolerance, and high-level abstractions for stateful serverless applications. The major novelity is its shared log design. The system is open-sourced at

The paper presents Cloudburst, a stateful FaaS platform that provides Python-like programming models. Cloudburst leverages Anna, a scalable key-value store, to achieve low-latency mutable state and communication.

Catalyzer proposes init-less, which leverages fork/snapshots to skip the initialization costs during startup, and can achieve <1ms startup latency in the best case. The system is based on virtualization based sandboxes, gVisor.

Faastlane aims to optimize the communication latency between functions, which is significant for function chain (or workflows in the title). The major idea is to use thread-level isolation, that means different functions are located in the same process, and Faastlane utilizes Intel MPK to provide isolation between different threads.

Replayable optimizes CRIU for docker containers to boost container startup latency and reduce memory costs.

SEUSS utilizes similir ideas as Catalyzer and Replayable to boost serverless latency (especially for startup latency) using fork/snapshots. The system is based on unikernels.

This paper presents Dorylus, a distributed system for training GNNs by utilizing serverless computing. Dorylus can achieve even better performance and low costs than CPU and GPU-based systems.

This paper presents Locus, a serverless analytics system combines cheap but slow storage with fast but expensive storage to balance performance and costs.

This paper presents Sprocket, a configurable, stage-based, scalable, serverless video processing framework that exploits intra-video parallelism to achieve low latency.

This paper presents an interesting serverless applications, which is a shell (called sshell), that allow users to run shell scripts on a serverless platform. The sourced code is avialable at