lockless-datastructures

I implemented an mpmc and spsc atomic ring buffer to learn about the memory models of modern cpu's As per my naive benchmarks the atomic ring buffer's turned out to be faster than the mutex based ring buffer

On the way even managed to write code which deadlocked(in the mutex ring buffer) and learned how mistakes are made!
This ring buffer's are used in high performance packet processing, the network directly writes into a ring buffer (dma) and the cpu picks them up when its free to.
I actually SAW the difference cache collisions make in performance (quite a lot btw)

Mutex Ring Buffer

Simple ring buffer with a mutex over the whole list (imho the source of perfomance difference)

Atomic Ring Buffer Spsc

Head and tail are declared as atomics, while reading and writing to them, I make sure appropriate memory fences are used.

Atomic Ring Buffer Mpmc

In this I maintain a sequence number is each slot, the writer checks if the sequence number is equal to head then only it starts writing. On finishing writing it increments the sequence number by 1. The sequence number are init to be equal to the index so,the reader looks if the sequence is equal to tail+1, then only does it read the slot.The reader on finishing reading makes the sequence number tail + N to make it ready for the next writer to write! All this additions are wrapped to handle even big numbers! The head and tail are modulo N ( I take N as a power of 2, so that I can do modulo with bit masking)

Use this?

This is supposed to be a library that I make use of in my upcoming projects!
You can clone this and add this as a crate to your project and use the AtomicRingBufferSpsc struct to init a new ring buffer.
Make sure the size is a power of 2, or it wont compile.
Boom now you have a ring buffer, with push and read,all your high performance networking dreams are fulfilled!!
The demo just demoing that my ring buffer works!