OneRingBuf

One Ring to rule them all, One Ring to find them,
One Ring to bring them all and in the mem'ry bind them
In the Land of Rust where the Data lies.

A lock-free, single-producer, single-consumer (SPSC) ring buffer optimised for real-time applications. It offers in-place mutability, asynchronous support, and virtual memory optimisation.

Key Features

• Lock-Free SPSC: Designed for high-throughput, low-latency communication between two threads.
• In-Place Mutability: Modify elements directly in the buffer without needing to move them, reducing copying and improving performance.
• Asynchronous Support: First-class async/await integration for non-blocking operations.
• Virtual Memory Optimisation: Uses vmem to map the buffer to contiguous virtual memory, allowing for safe and efficient access to slices of data.
• Flexible Memory Allocation: Supports both stack and heap allocation to suit different needs.

When to Use oneringbuf

This crate was specifically developed for real-time audio stream processing, but it is well-suited for any scenario that requires a high-performance SPSC queue, such as:

• Real-time signal processing
• Game development (e.g. for audio or event queues)
• High-frequency data streaming
• Anywhere you need to pass data between two threads with minimal overhead.

A practical example for audio processing can be found here.

Getting Started

Add oneringbuf to your Cargo.toml:

[dependencies]
oneringbuf = "0.6.0" # Replace with the latest version

Here is a basic example of creating a buffer, producing, and consuming data:

use oneringbuf::LocalHeapRB;

fn main() {
    // Create a heap-allocated ring buffer with a capacity for 1024 elements.
    let buf = LocalHeapRB::from(vec![0; 1024]);
    let (mut prod, mut cons) = buf.split();

    // Push some data into the buffer.
    for i in 1..=10 {
        prod.push(i).unwrap();
    }

    // Consume the data from the buffer.
    for _ in 0..10 {
        if let Some(val) = cons.pop() {
            println!("Popped: {}", val);
        }
    }
}

Choosing the Right Buffer

oneringbuf provides several buffer types to suit different use cases. Here’s a guide to help you choose:

• Stack-Allocated (*StackRB): Ideal for small buffers where the size is known at compile time. Offers fast allocation but is limited by stack space.
• Heap-Allocated (*HeapRB): Suitable for large buffers or when the size is determined at runtime. More flexible but with a small overhead for dynamic memory management.

And for concurrency:

• Local (Local*RB): For single-threaded use. Offers the best performance due to simpler synchronisation.
• Shared (Shared*RB): For multi-threaded environments where the producer and consumer are on different threads. Uses atomic operations for thread safety.
• Async (Async*RB): Designed for async/await, providing non-blocking operations suitable for asynchronous runtimes.

Core Concepts

In-place Mutability

A key feature of oneringbuf is the ability to modify elements directly within the buffer. This is achieved through the WorkIter, which provides mutable references to items that have been produced but not yet consumed. This avoids unnecessary data copying and can significantly improve performance in scenarios like audio processing or data transformations.

use oneringbuf::{LocalHeapRBMut, ORBIterator};

let buf = LocalHeapRBMut::from(vec![0; 4096]);
let (mut prod, mut work, mut cons) = buf.split_mut();

// Produce some data
prod.push(10).unwrap();

// Mutate the data in-place
if let Some(item) = work.get_mut() {
    *item *= 2;
    unsafe { work.advance(1); }
}

// The consumer will now see the modified value
assert_eq!(cons.pop(), Some(20));

Advanced Features

Virtual Memory (vmem) Optimisation

For circular buffers, a powerful optimisation is to map the underlying buffer to two contiguous regions of virtual memory. This allows you to treat the circular buffer as a linear one, making it possible to read or write data across the wrap-around point in a single operation. More information can be found here.

This crate supports this optimisation through the vmem feature flag, which is currently limited to unix targets. When using vmem, the buffer size (length of the buffer times the size of the stored type) must be a multiple of the system's page size (usually 4096 for x86_64).

At the moment, the feature has been tested on GNU/Linux, Android, macOS and iOS.

A Note About iOS

vmem works by allocating shared memory. While this doesn't represent a problem on other platforms, it is different on iOS. Users should create an app group (more information here) and then set the environment variable IOS_APP_GROUP_NAME to the name of that group.

Building and Running Examples

To run the tests, benchmarks, or examples, clone the repository and use the following commands from the root directory.

To run tests:

cargo +nightly test --tests

To run benchmarks:

cargo bench

To run the CPAL example:

RUSTFLAGS="--cfg cpal" cargo run --example cpal

If you encounter an error like ALSA lib pcm_dsnoop.c:567:(snd_pcm_dsnoop_open) unable to open slave, please refer to this issue.

To run any other example (e.g., simple_async):

cargo run --example simple_async --features async

Best Practices

1. Handle Uninitialised Items with Care: If you create a buffer with uninitialised memory (e.g., with new_zeroed), you must use *_init methods (e.g., push_init) to safely write data. Using normal methods on uninitialised memory can lead to Undefined Behaviour.
2. Match Buffer Type to Use Case: Choose the buffer type that best fits your performance and concurrency needs.
3. Buffer Sizing for vmem: When using the vmem feature, ensure your buffer size is a multiple of the system's page size to prevent panics.
4. Graceful Shutdown: The buffer is deallocated only when the last of its iterators is dropped. Ensure all iterators are dropped for proper cleanup.

License

This work is dual-licensed under:

• Apache License, Version 2.0 LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0
• MIT license LICENSE-MIT or https://opensource.org/license/MIT

You can choose between one of them.