GeoFF: Federated Serverless Workflows with Data Pre-Fetching

Overview

• Function-as-a-Service (FaaS) is a cloud computing model where applications are implemented as workflows of independent functions.
• Cloud providers offer composition services for these workflows, but they do not support cross-platform workflows, forcing developers to hardcode composition logic.
• FaaS workflows can be slow due to issues like cascading cold starts, inter-function latency, and data download latency.

Plain English Explanation

The paper proposes a solution called GeoFF, a serverless choreography middleware that can execute FaaS workflows across different public and private FaaS platforms. GeoFF supports ad-hoc workflow recomposition, which means it can dynamically adjust the workflow as needed without hard-coding the composition logic.

To address the performance issues of FaaS workflows, GeoFF also supports function pre-warming and data pre-fetching. This helps minimize end-to-end workflow latency by taking the cold starts and data download latency off the critical path. In other words, GeoFF tries to pre-load the necessary functions and data ahead of time so that when the workflow is executed, it runs more smoothly without those delays.

Technical Explanation

The paper presents the design and implementation of GeoFF, a serverless choreography middleware that can execute FaaS workflows across different platforms. GeoFF provides ad-hoc workflow recomposition, which allows it to dynamically adjust the workflow without hardcoding the composition logic.

To improve performance, GeoFF supports function pre-warming and data pre-fetching. Function pre-warming keeps functions "warm" and ready to execute, avoiding the cold start latency. Data pre-fetching downloads necessary data in advance, reducing the download latency on the critical path.

The paper evaluates GeoFF's performance using a proof-of-concept prototype and a realistic application. The results show that GeoFF can reduce end-to-end workflow latency by more than 50% compared to traditional FaaS workflows.

Critical Analysis

The paper addresses an important challenge in the FaaS space – the lack of support for cross-platform workflow execution and the performance issues caused by cold starts and data download latency.

One potential limitation is that the paper only evaluates GeoFF's performance using a single realistic application. It would be helpful to see how GeoFF performs with a wider range of applications and workloads to better understand its generalizability.

Additionally, the paper does not discuss the overhead or complexity introduced by GeoFF's workflow recomposition and pre-warming/pre-fetching mechanisms. It would be valuable to understand the tradeoffs and potential downsides of these features.

Overall, the paper presents a promising approach to address the challenges of FaaS workflows, but further research and evaluation would be needed to fully assess its practical implications and limitations.

Conclusion

The paper introduces GeoFF, a serverless choreography middleware that can execute FaaS workflows across different platforms and optimizes performance through function pre-warming and data pre-fetching. This helps address the limitations of current FaaS platforms, which do not support cross-platform workflows and suffer from performance issues.

The experimental results show that GeoFF can significantly reduce the end-to-end latency of FaaS workflows, potentially making serverless computing more viable for a wider range of applications. While further research is needed, this work represents an important step towards more robust and efficient FaaS platforms.