TL;DR
This report summarizes the outcomes of a Dagstuhl Seminar on Serverless Computing, aiming to synthesize a shared vision for the next decade. It identifies key challenges and opportunities across design, software engineering, applications, and evaluation, emphasizing the need for interdisciplinary collaboration to establish serverless as a mature and viable technology.
What Problem Does This Paper Address?
The paper addresses the challenges and opportunities presented by the rapid rise of serverless computing. While serverless offers increased accessibility, fine-grained control, and affordability for cloud services, it also suffers from a scattered vision, a lack of coherent design approaches, and significant technical complexity. The authors highlight the need to bridge the gap between industry adoption and academic research to establish serverless computing as a robust and sustainable technology for the future. This involves tackling issues related to system design, software engineering practices, understanding diverse application requirements, and developing effective evaluation methodologies.
Methods
This paper is not a traditional research paper presenting novel algorithms or experiments. Instead, it is a report on the outcomes of a Dagstuhl Seminar. The "methods" employed were those of a collaborative workshop:
- Bringing together diverse experts: Researchers, developers, and practitioners from software engineering, parallel and distributed systems, and performance analysis were convened.
- Focused discussions: The seminar was structured around four key topics: design decisions, software engineering, applications and requirements, and evaluation.
- Keynotes and working groups: Distinguished speakers presented on specific aspects, followed by in-depth discussions within working groups.
- Synthesis of vision: The report synthesizes the collective insights, identifying open challenges, proposing future research directions, and aiming to establish a shared vision for the next decade of serverless computing.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Serverless is a "cloud continuum" enabler: Serverless computing is seen as a key technology for unifying resources from datacenters to the edge, offering a path towards a more integrated and accessible computing landscape.
- The "Serverless End Game": A vision of full transparency and near-zero operational concerns for users is proposed, where computing resources become standardized utilities, making distributed programming more democratized.
- Developer Experience is Crucial: The ease of use, tooling, and architectural patterns significantly impact developer adoption and productivity in serverless environments.
- Reproducibility is a Major Hurdle: Similar to other areas of computer science, achieving reproducible results in serverless performance evaluation is a significant challenge due to the opaque nature of underlying systems and rapid evolution.
- Beyond FaaS: While Function-as-a-Service (FaaS) is a prominent example, the broader serverless computing paradigm also encompasses Backend-as-a-Service (BaaS) and other specialized components, forming a complex ecosystem.
- Interdisciplinary Collaboration is Essential: Addressing the multifaceted challenges of serverless computing requires strong collaboration between computer systems, software engineering, and performance engineering communities.
Key Results
This paper does not present quantitative results in the traditional sense. Instead, its key "results" are the synthesized insights and identified challenges from the Dagstuhl Seminar. These include:
- A proposed reference architecture for serverless computing: This architecture outlines key layers from hardware to programming models and toolchains, emphasizing the ecosystem nature of serverless.
- Identification of critical challenges: The report details numerous challenges across the software engineering lifecycle (planning, design, implementation, testing, deployment), system-level operational requirements, programming models, workload management, and ecosystem stability.
- A definition of serverless computing: A consensus definition is provided, highlighting key characteristics like NoOps, utilization-based billing, and offloaded operational concerns.
- Identification of key application domains: Scientific computing, machine learning, online gaming, and mobile/telco are highlighted as domains with significant potential and unique challenges for serverless adoption.
Limitations and Future Work
- Infancy of the Paradigm: Serverless computing is acknowledged to be in its early stages, with many layers (runtime engines, toolchains) still maturing and potentially not converging.
- Vendor Lock-in: The proprietary nature of many serverless offerings leads to compatibility issues and vendor lock-in, hindering migration and competition.
- Need for Better Tooling: Immature tooling for infrastructure-as-code (IaC), debugging, and observability remains a significant bottleneck.
- Reproducibility: The challenge of reproducible performance evaluation needs continued focus.
- Future Work: The authors suggest several next steps for the community, including:
- Completing the transition to "Serverless 2.0" by measuring and optimizing prototypes.
- Understanding the cost and economics of serverless.
- Conducting more empirical research, especially on failures.
- Supporting serverless functions that can co-exist and meet diverse requirements.
- Investigating the "greenness" and cost efficiency of serverless.
- Co-designing forward-looking serverless architectures for heterogeneous environments.
- Establishing a dedicated conference series on Serverless Software and Systems.
Summary updated