ZXC | Not Hacker News!

Discussion (1 comments)

Showing 6 comments

13 days ago

2 replies

> Unlike symmetric codecs (LZ4), ZXC trades compression speed (build-time) for maximum decompression throughput (run-time).

Huh? This is exactly what lz4 does. Yes, zxc is faster than lz4 at decompressing, but lz4 is clearly asymmetric in the benchmarks shown further down on the page.

pollop_

12 days ago

1 reply

Fair point regarding the terminology. You are right that LZ4 is already asymmetric compared to generic compression.

My phrasing was a bit too binary. What I meant to highlight is that ZXC pushes this trade-off much further than LZ4. I designed ZXC specifically for scenarios where we accept a significantly slower build-time/compression cost in exchange for that extra raw decode throughput.

aidenn0

12 days ago

I think the existence of lz4hc is evidence that what you are discussing is within the design-space of LZ4, but ZXC just did a better job of it (though lz4hc achieves better compression than ZXC)

masklinn

13 days ago

And afaik this is what most compression algorithms do. It’s certainly one of the trade offs you have to consider when archiving at high levels of compression.

pollop_

17 days ago

Hello HN, author here.

I built ZXC, a lossless compression library designed specifically for "Write Once, Read Many" scenarios (Game assets, Firmware, App bundles).

The Concept: Standard real-time codecs (LZ4, Snappy) are designed for symmetry. ZXC explores an asymmetric approach: it spends significantly more CPU cycles at compression time to generate a bitstream specifically structured for modern, wide-issue execution pipelines (maximizing Instruction Level Parallelism).

Performance (Verified via lzbench integration): ZXC was merged into the official lzbench repository monday.

ARM64 (Apple M2): 6,365 MB/s decoding (~40% faster than LZ4). x86_64 (AMD EPYC 9554): At Level 5, ZXC achieves compression ratios comparable to zstd --fast while maintaining nearly 2x the decompression speed (3.8 GB/s vs 2.0 GB/s).

Implementation Details: Written in C99, zero internal allocation (caller provides buffers). Designed for easy FFI (Go/Python/Rust bindings). Validated via continuous fuzzing (ClusterFuzzLite) and static analysis. PR for OSS-Fuzz integration is ready.

This is a v0.1.1 release. The bitstream format is stable but open to improvements based on feedback from the community.

https://github.com/hellobertrand/zxc

sounds

13 days ago

If you want to see a comparison against an even broader set of open source compression algos, this is lzbench (it's linked directly from the ZXC github page)

lzbench has added ZXC to its suite. This makes a nice apples to apples comparison possible.

https://github.com/inikep/lzbench

Resources