Fountain code which is an order of magnitude faster than RaptorQ.

cberner / raptorq

Rust implementation of RaptorQ (RFC6330)

Apache License 2.0

264 stars 47 forks source link

Fountain code which is an order of magnitude faster than RaptorQ. #142

Closed fadedbee closed 1 year ago

fadedbee commented 1 year ago

I was looking at your RaptorQ implementation to evaluate the speed of an original fountain code which I have developed. (It's mathematically original, not based on RS, LT or tornado codes.)

On encoding (with a small number of symbols) I see an order of magnitude better performance.

Do you have any significant optimisations which may make this RaptorQ implementation faster? i.e. Is it already representative of the state of the art?
Are you interested in following this work? (I need to code equivalent benchmarks and create a file-focused command-line fountain and combiner.) I couldn't see your email address on github.

fadedbee commented 1 year ago

On my 2500K i5:

     Running benches/encode_benchmark.rs (target/release/deps/encode_benchmark-11ec8998e3d558b6)
Symbol size: 1280 bytes (without pre-built plan)
symbol count = 10, encoded 127 MB in 0.807secs, throughput: 1268.8Mbit/s
symbol count = 100, encoded 127 MB in 0.645secs, throughput: 1586.7Mbit/s
symbol count = 250, encoded 127 MB in 0.659secs, throughput: 1552.3Mbit/s

    Finished release [optimized] target(s) in 0.52s
     Running `target/release/fountain`
992.984852ms to encode 1250MB (10Gbit)

cberner commented 1 year ago

I've heard that there are optimizations to make RaptorQ ~5x faster. However, they are proprietary and I don't know what they are, so I haven't been able to implement them.

I'm no longer working on fountain codes, but sure when you finish it feel free to post the results here. I'd be interested

cberner commented 1 year ago

Closing for now, but feel free to re-open when you have results.