isa-l VS rapidgzip

For using accelerators (QAT, I assume, because the article is about this) from the CLI, then QATzip[3] comes with a command line tool qatzip which can be used as described in the project ReadMe. I didn't test it, though, as I have no QAT-enabled device.

[1] https://github.com/intel/isa-l/tree/master/igzip

Ex 1: https://github.com/intel/isa-l/blob/master/README.md

Examples: - Someone I know created machine-optimized library in c and asm to accelerate compression, erasure coding, and encryption on Intel Xeon CPUs (https://github.com/intel/isa-l) which runs those ops WAY faster than the standard linux libraries. - My friend is making a retro SEGA genesis game using a C and asm dev kit (https://github.com/Stephane-D/SGDK)

igzip (https://github.com/intel/isa-l) is much faster than gzip or pigz when it comes to decompression, 2-3x in my experience. There is also a Python module (isal) that provides a GzipFile-like wrapper class, for an easy speed-up of Python scripts that read gzipped files.

However, it only supports up to level 3 when compressing data, so it can't be used as a drop-in replacement for gzip. You also need to make sure to use the latest version if you are going to use it in the context of bioinformatics, since older versions choke on concatenated gzip files common in that field.

I also did benchmarks with zlib and libarchivemount via their library interface here [0]. It has been a while that I have run them, so I forgot. Unfortunately, I did not add libdeflate.

[0] https://github.com/mxmlnkn/rapidgzip/blob/master/src/benchma...

Hi, author here.

You are right in the index being the easy-mode. Over the years there have been lots of implementations trying to add an index like that to the gzip metadata itself or as a sidecar file, with bgzip probably being the most known one. None of them really did stick, hence the necessity for some generic multi-threaded decompressor. A probably incomplete list of such implementations can be found in this issue: https://github.com/mxmlnkn/rapidgzip/issues/8

The index makes it so easy that I can simply delegate decompression to zlib. And since paper publication I've actually improved upon this by delegating to ISA-l / igzip instead, which is twice as fast. This is already in the 0.8.0 release.

As derived from table 1, the false positive rate is 1 Tbit / 202 = 5 Gbit or 625 MB for deflate blocks with dynamic Huffman code. For non-compressed blocks, the false positive rate is roughly one per 500 KB, however non-compressed blocks can basically be memcpied or skipped over and then the next deflate header can be checked without much latency. On the other hand, for dynamic blocks, the whole block needs to be decompressed first to find the next one. So the much higher false positive rate for non-compressed blocks doesn't introduce that much overhead.

I have some profiling built into rapidgzip, which is printed with -v, e.g., rapidgzip -v -d -o /dev/null 20xsilesia.tar.gz :

    Time spent in block finder              : 0.227751 s

I have not only implemented parallel decompression but also random access to offsets in the stream with https://github.com/mxmlnkn/pragzip I did some benchmarks on some really beefy machines with 128 cores and was able to reach almost 20 GB/s decompression bandwidth. The single-core decoder has lots of potential for optimization because I had to write it from scratch, though.

Decompression of arbitrary gzip files can be parallelized with pragzip: https://github.com/mxmlnkn/pragzip

At the very least you are duplicating logic without the exception. The check for eof has to be done implicitly anyway inside read because it has to fill the bit buffer with data from the byte buffer or the byte buffer with data from the file. And if both fail, then we already know the result of eof, so no need to duplicate checking for eof in the outer read calling loop.

Here is the full commit with ad-hoc benchmark results in the commit message:

https://github.com/mxmlnkn/pragzip/commit/0b1af498377838c30f...

and here the benchmarks I ran at that time:

https://github.com/mxmlnkn/pragzip/blob/0b1af498377838c30fea...

As you can see, it's part of my random-seekable multi-threaded gzip and bzip2 parallel decompression libraries.

What you can also see in the commit message is that it wasn't a 50% time reduction but a 50% bandwidth increase, which would translate to a 30% time reduction. It seems I remembered that partly wrong. But it still was a significant optimization for me.