This talk will delve deep into the innovative use of threads versus fibers in user-space cooperative scheduling, drawing upon real-world applications in the context of planet-scale storage systems, particularly Google Cloud Storage.
This talk will explore how Google Cloud Storage leverages fibers to address the challenge of "noisy neighbors" – non-conformant clients that can monopolize resources and degrade performance for others. By employing fibers and user-space cooperative scheduling, GCS achieves fairer CPU resource allocation, prioritizing conformant clients while strategically throttling non-conformant ones.

During this session, I will explore the architectural considerations and performance trade-offs between threads and fibers, leveraging my experiences from leading this project at Google.
This talk will be particularly beneficial for engineers and architects involved in building or maintaining large-scale distributed systems.


In the dynamic landscape of cloud computing, efficient task management is crucial for optimizing resource utilization and maintaining application performance. This talk will delve deep into the innovative use of threads versus fibers in user-space cooperative scheduling, drawing upon real-world applications in the context of planet-scale storage systems, particularly Google Cloud Storage.

Threads, though prevalent in multithreading, often suffer from context switching overhead and potential interference, commonly known as the 'noisy neighbor' problem, impacting CPU isolation. Fibers, on the other hand, offer a lightweight alternative by facilitating task execution within the same thread, reducing overhead and enhancing system responsiveness.

This talk will explore how Google Cloud Storage leverages fibers to address the challenge of "noisy neighbors" – non-conformant clients that can monopolize resources and degrade performance for others. By employing fibers and user-space cooperative scheduling, GCS achieves fairer CPU resource allocation, prioritizing conformant clients while strategically throttling non-conformant ones.

During this session, I will explore the architectural considerations and performance trade-offs between threads and fibers, leveraging my experiences from leading projects at Google. Attendees will gain insights into how fibers can be employed to enhance scheduling efficiency in cloud environments, ensuring optimal CPU isolation and minimal latency.

Key challenges tackled include designing isolation mechanisms that prevent resource contention, thereby improving SLAs for high-demand cloud storage clients. Moreover, I'll showcase strategies that integrate fibers seamlessly into existing multithreaded systems.

This talk will be particularly beneficial for engineers and architects involved in building or maintaining large-scale distributed systems. Attendees can expect to leave with a solid understanding of how user-space cooperative scheduling using fibers can be leveraged to address contemporary challenges in cloud storage, potentially driving down costs and boosting application performance across distributed infrastructures.