Turns out that these numbers are caused by APFS weirdness. I used 'du' to get them which reports the size on disk, which is weirdly bloated for some reason when compressing in parallel. I should've used 'du -A', which reports the apparent size.

Here's a table with the correct sizes, reported by 'du -A' (which shows the apparent size):

    +---------+---------+--------+--------+--------+
    |  none   |  zstd   |   xz   |  gzip  | brotli |
    +---------|---------|--------|--------|--------|
    | 47.81M  | 37.92M  | 37.96M | 38.80M | 37.06M |
    +---------+---------+--------+--------+--------+

These numbers are much more impressive. Still, Brotli has a slight edge.

> | 1.1M | 2.0M | 1.1M | 1.1M | 1.1M |

Something is going terribly wrong with `zstd` here, where it is reported to compress a file of 1.1MB to 2MB. Zstd should never grow the file size by more than a very small percent, like any compressor. Am I interpreting it correctly that you're doing something like `zstd -22 --ultra $FILE && wc -c $FILE.zst`?

If you can reproduce this behavior, can you please file an issue with the zstd version you are using, the commands used, and if possible the file producing this result.

Okay now this is weird.

I can reproduce it just fine ... but only when compressing all PDFs simultaneously.

To utilize all cores, I ran:

    $ for x in *.pdf; do zstd <"$x" >"$x.zst" --ultra -22 & done; wait

(and similar for the other formats).

I ran this again and it produced the same 2M file from the source 1.1M file. However when I run without paralellization:

    $ for x in *.pdf; do zstd <"$x" >"$x.zst" --ultra -22; done

That one file becomes 1.1M, and the total size of *.zst is 37M (competitive with Brotli, which is impressive given how much faster it is to decompress).

What's going on here? Surely '-22' disables any adaptive compression stuff based on system resource availability and just uses compression level 22?

Yeah, `--adaptive` will enable adaptive compression, but it isn't enabled by default, so shouldn't apply here. But even with `--adaptive`, after compressing each block of 128KB of data, zstd checks that the output size is < 128KB. If it isn't, it emits an uncompressed block that is 128KB + 3B.

So it is very central to zstd that it will never emit a block that is larger than 128KB+3B.

I will try to reproduce, but I suspect that there is something unrelated to zstd going on.

What version of zstd are you using?

'zstd --version' reports: "** Zstandard CLI (64-bit) v1.5.7, by Yann Collet **". This is zstd installed through Homebrew on macOS 26 on an M1 Pro laptop. Also of interest, I was able to reproduce this with a random binary I had in /bin: https://floss.social/@mort/115940378643840495

I was completely unable to reproduce it on my Linux desktop though: https://floss.social/@mort/115940627269799738

I've figured out the issue. Use `wc -c` instead of `du`.

I can repro on my Mac with these steps with either `zstd` or `gzip`:

    $ rm -f ksh.zst
    $ zstd < /bin/ksh > ksh.zst
    $ du -h ksh.zst
    1.2M ksh.zst
    $ wc -c ksh.zst
     1240701 ksh.zst
    $ zstd < /bin/ksh > ksh.zst
    $ du -h ksh.zst
    2.0M ksh.zst
    $ wc -c ksh.zst
     1240701 ksh.zst
    
    $ rm -f ksh.gz
    $ gzip < /bin/ksh > ksh.gz
    $ du -h ksh.gz
    1.2M ksh.gz
    $ wc -c ksh.gz
     1246815 ksh.gz
    $ gzip < /bin/ksh > ksh.gz
    $ du -h ksh.gz
    2.1M ksh.gz
    $ wc -c ksh.gz
     1246815 ksh.gz

When a file is overwritten, the on-disk size is bigger. I don't know why. But you must have ran zstd's benchmark twice, and every other compressor's benchmark once.

I'm a zstd developer, so I have a vested interest in accurate benchmarks, and finding & fixing issues :)

Interesting!

It doesn't seem to be only about overwriting, I can be in a directory without any .zst files and run the command to compress 55 files in parallel and it's still 45M according to 'du -h'. But you're right, 'wc -c' shows 38809999 bytes regardless of whether 'du -h' shows 45M after a parallel compression or 38M after a sequential compression.

My mental model of 'du' was basically that it gives a size accurate to the nearest 4k block, which is usually accurate enough. Seems I have to reconsider. Too bad there's no standard alternative which has the interface of 'du' but with byte-accurate file sizes...

Yeah, it isn't quite that simple. E.g. `/bin/ksh` reports 1.4MB, but it is actually 2.4MB. Initially, I thought it was because the file was sparse, but there are only 493KB of zeros. So something else is going on. Perhaps some filesystem-level blocks are deduped from other files? Or APFS has transparent compression? I'm not sure.

It does still seem odd that APFS is reporting a significantly larger disk-size for these files. I'm not sure why that would ever be the case, unless there is something like deferred cleanup work.

Ross Burton on Mastodon suggests that it might be deduplication; when writing sequentially, later files can re-use blocks from earlier files, while that isn't the case as much when writing sequentially. That seems plausible enough to me.

I've concluded that this can't be the reason. It'd only result in an error where the size reported by 'du' is smaller than the apparent size (aka number of bytes reported by 'wc -c') of the file. What we see here is that the size reported by 'du' is almost twice as large as the number of bytes. That can't be the result of dedpulication.

I'll chalk it up to "some APFS weirdness".

doesn't zstd cap out at compression level 19?

From the man page:

    --ultra: unlocks high compression levels 20+ (maximum 22), using a lot more memory.

Regardless, this reproduces with random other files and with '-9' as the compression level. I made a mastodon post about it here: https://floss.social/@mort/115940378643840495

If you're worried about double-compression of image data, you can uncompress all images by using qpdf:

    qpdf --stream-data=uncompress in.pdf out.pdf

The resulting file should compress better with zstd.

Why not use a more widespread compression algorithm (e.g. gzip) considering that Brotli barely performs better at all? Sounds like a pain for portability

I'm not sold on the idea of adding compression to PDF at all, I'm not convinced that the space savings are worth breaking compatibility with older readers. Especially when you consider that you can just compress it in transit with e.g HTTP's 'Content-Encoding' without any special PDF reader support. (You can even use 'Content-Encoding: br' for brotli!)

If you do wanna change PDF backwards-incompatibly, I don't think there's a significant advantage to choosing gzip to be honest, both brotli and zstd are pretty widely available these days and should be fairly easy to vendor. But yeah, it's a slight advantage I guess. Though I would expect that there are other PDF data sets where brotli has a larger advantage compared to gzip.

But what I really don't get is all the calls to use zstd instead of brotli and treating the choise to use brotli instead of zstd as some form of Google conspiracy. (Is Facebook really better?)

>But what I really don't get is all the calls to use zstd instead of brotli and treating the choise to use brotli instead of zstd as some form of Google conspiracy. (Is Facebook really better?)

I may dislike Google. But my support of JPEG XL and Zstd has nothing to do with competition tech being Google at all. I simply think JPEG XL and Zstd are better technology.

Could you add compression and decompression speeds to your table?

I just did some interactive shell loops and globs to compress everything and output CSV which I processed into an ASCII table, so I don't exactly have a pipeline I can modify and re-run the tests with compression speeds added ... but I can run some more interactive shell-glob-and-loop-based analysis to give you decompression speeds:

    ~/tmp/pdfbench $ hyperfine --warmup 2 \
    'for x in zst/*; do zstd -d >/dev/null <"$x"; done' \
    'for x in gz/*; do gzip -d >/dev/null <"$x"; done' \
    'for x in xz/*; do xz -d >/dev/null <"$x"; done' \
    'for x in br/*; do brotli -d >/dev/null <"$x"; done'
    Benchmark 1: for x in zst/*; do zstd -d >/dev/null <"$x"; done
      Time (mean ± σ):     164.6 ms ±   1.3 ms    [User: 83.6 ms, System: 72.4 ms]
      Range (min … max):   162.0 ms … 166.9 ms    17 runs
    
    Benchmark 2: for x in gz/*; do gzip -d >/dev/null <"$x"; done
      Time (mean ± σ):     143.0 ms ±   1.0 ms    [User: 87.6 ms, System: 43.6 ms]
      Range (min … max):   141.4 ms … 145.6 ms    20 runs
    
    Benchmark 3: for x in xz/*; do xz -d >/dev/null <"$x"; done
      Time (mean ± σ):     981.7 ms ±   1.6 ms    [User: 891.5 ms, System: 93.0 ms]
      Range (min … max):   978.7 ms … 984.3 ms    10 runs
    
    Benchmark 4: for x in br/*; do brotli -d >/dev/null <"$x"; done
      Time (mean ± σ):     254.5 ms ±   2.5 ms    [User: 172.9 ms, System: 67.4 ms]
      Range (min … max):   252.3 ms … 260.5 ms    11 runs
    
    Summary
      for x in gz/*; do gzip -d >/dev/null <"$x"; done ran
        1.15 ± 0.01 times faster than for x in zst/*; do zstd -d >/dev/null <"$x"; done
        1.78 ± 0.02 times faster than for x in br/*; do brotli -d >/dev/null <"$x"; done
        6.87 ± 0.05 times faster than for x in xz/*; do xz -d >/dev/null <"$x"; done

As expected, xz is super slow. Gzip is fastest, zstd being somewhat slower, brotli slower again but still much faster than xz.

    +-------+-------+--------+-------+
    | gzip  | zstd  | brotli | xz    |
    +-------+-------+--------+-------+
    | 143ms | 165ms | 255ms  | 982ms |
    +-------+-------+--------+-------+

I honestly expected zstd to win here.

Thanks a lot. Interestingly Brotli’s author mentioned here that zstd is 2× faster at decompressing, which roughly matches your numbers:

https://news.ycombinator.com/item?id=46035817

I’m also really surprised that gzip performs better here. Is there some kind of hardware acceleration or the like?

Zstd should not be slower than gzip to decompress here. Given that it has inflated the files to be bigger than the uncompressed data, it has to do more work to decompress. This seems like a bug, or somehow measuring the wrong thing, and not the expected behavior.

It seems like zstd is somehow compressing really badly when many zstd processes are run in parallel, but works as expected when run sequentially: https://news.ycombinator.com/item?id=46723158

Regardless, this does not make a significant difference. I ran hyperfine again against a 37M folder of .pdf.zst files, and the results are virtually identical for zstd and gzip:

    +-------+-------+--------+-------+
    | gzip  | zstd  | brotli | xz    |
    +-------+-------+--------+-------+
    | 142ms | 165ms | 269ms  | 994ms |
    +-------+-------+--------+-------+

Raw hyperfine output:

    ~/tmp/pdfbench $ du -h zst2 gz xz br
     37M    zst2
     38M    gz
     38M    xz
     37M    br
    
    ~/tmp/pdfbench $ hyperfine ...
    Benchmark 1: for x in zst2/*; do zstd -d >/dev/null <"$x"; done
      Time (mean ± σ):     164.5 ms ±   2.3 ms    [User: 83.5 ms, System: 72.3 ms]
      Range (min … max):   162.3 ms … 172.3 ms    17 runs
    
    Benchmark 2: for x in gz/*; do gzip -d >/dev/null <"$x"; done
      Time (mean ± σ):     142.2 ms ±   0.9 ms    [User: 87.4 ms, System: 43.1 ms]
      Range (min … max):   140.8 ms … 143.9 ms    20 runs
    
    Benchmark 3: for x in xz/*; do xz -d >/dev/null <"$x"; done
      Time (mean ± σ):     993.9 ms ±   9.2 ms    [User: 896.7 ms, System: 99.1 ms]
      Range (min … max):   981.4 ms … 1007.2 ms    10 runs
    
    Benchmark 4: for x in br/*; do brotli -d >/dev/null <"$x"; done
      Time (mean ± σ):     269.1 ms ±   8.8 ms    [User: 176.6 ms, System: 75.8 ms]
      Range (min … max):   261.8 ms … 287.6 ms    10 runs

Ah I understand. In this benchmark, Zstd's decompression time is 284 MB/s, and Gzip's is 330 MB/s. This benchmark is likely dominated by file IO for the faster decompressors.

On the incompressible files, I'd expect decompression of any algorithm to approach the speed of `memcpy()`. And would generally expect zstd's decompression speed to be faster. For example, on a x86 core running at 2GHz, Zstd is decompressing a file at 660 MB/s, and on my M1 at 1276 MB/s.

You could measure locally either using a specialized tool like lzbench [0], or for zstd by just running `zstd -b22 --ultra /path/to/file`, which will print the compression ratio, compression speed, and decompression speed.

[0] https://github.com/inikep/lzbench