Thanks Antoine,

I will work with Kosta to update the numbers for the post. In the interim,
in case anyone is interested, you can see early results here[1]. Spoiler is
that  BYTE_STREAM_SPLIT + ZSTD
* Gets the best average compression ratio
* Same average compression speed as PLAIN+ZSTD (both about 2x faster than
ALP)
* An order of magnitude slower for decompression and random lookup than ALP

Andrew

[1]: https://github.com/alamb/alp_benchmark/issues/3

On Tue, Aug 25, 2026 at 4:44 AM Antoine Pitrou <[email protected]> wrote:

>
> It would be nice if the post included BYTE_STREAM_SPLIT + ZSTD in the
> benchmark comparisons.
>
> Regards
>
> Antoine.
>
>
> Le 23/08/2026 à 17:17, PRATEEK GAUR a écrit :
> > Enjoyed working with Kosta and Andrew on this.
> >
> > I was waiting for the email to show up in my inbox after Kosta sent it.
> >
> > But then realized this might have gone to spam.
> > Hoping replying to the email thread will pull it out of spam for others.
> >
> > Best
> > Prateek
> >
> > On Sat, Aug 22, 2026 at 9:22 AM Kosta Tarasov <
> [email protected]>
> > wrote:
> >
> >> Hi everyone,
> >>
> >> Andrew, Prateek, and I have been working on the ALP blog[1] and we think
> >> it's ready for review/comment.
> >>
> >> The blog covers the motivation for ALP, performance results, a technical
> >> overview, and ecosystem adoption.
> >>
> >> A rendered preview is available[2].
> >>
> >> Thanks,
> >> Kosta
> >>
> >> [1]: https://github.com/apache/parquet-site/pull/195
> >>
> >> [2]:
> >>
> https://alamb.github.io/parquet-site/blog/2026/08/14/alp-adaptive-lossless-floating-point-encoding-in-apache-parquet/
> >>
> >
>
>
>

Reply via email to