Folo

Mechanisms for high-performance hardware-aware programming in Rust.

The design tenets this project aims to satisfy are the following:

• In services, keep the processing of each request on a single processor to ensure both that data is locally cached for fast access and to avoid polluting caches of many processors with data of a single request.
• Be aware of memory region boundaries when scheduling work and placing data. Avoid moving data across these boundaries because it can be very slow.
• Use single-threaded logic without synchronization - even atomics and "lock-free" synchronization primitives are expensive compared to single-threaded logic. Whenever feasible, use !Send types to avoid accidental multithreading. Maintain separate mutable data sets per thread or memory region instead of maintaining global data sets.
• Use asynchronous logic in the app, in library code and when communicating with the operating system, ensuring that a thread is never blocked from doing useful work.

Contents

This is an umbrella project that covers multiple largely independent crates:

Some auxiliary crates are also published because the primary crates above require their functionality. They only indirectly contribute to the Folo mission, so are listed separately:

There are also some development-only crates in this repo, which are not published:

Development environment setup

See DEVELOPMENT.md.