You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@arrow.apache.org by Andy Grove <ag...@apache.org> on 2021/08/08 20:29:36 UTC
[Rust] Ballista status and benchmarks
I wanted to share a quick update on the status of Ballista.
Ballista is now capable of running some of the TPC-H benchmark queries
against the 1TB data set. I documented the benchmark results for
DataFusion, Ballista, and Apache Spark for reference, here:
https://github.com/andygrove/ballista-research/wiki/Ballista-benchmarks-results
Here is a chart showing these initial results. As noted on the wiki page, I
didn't put too much thought into the configurations used but they are at
least documented. I now plan on iterating on my benchmark process to
measure memory usage and show scalability as executors/cores are added for
each solution.
[image: sf1kcsv.png]
Note that I hit failures with all three solutions and this is why there are
columns missing from the chart.
I'll try and get something more compelling written up for a blog post to
coincide with the upcoming release of Ballista 0.5.0 but I figured folks
might be interested in these informal interim results.
Thanks,
Andy.
Re: [Rust] Ballista status and benchmarks
Posted by QP Hou <ho...@gmail.com>.
I am not able to load the benchmark graph png, so reattaching it here
in case other people have the same issue.
On Sun, Aug 8, 2021 at 1:31 PM Andy Grove <ag...@apache.org> wrote:
>
> I wanted to share a quick update on the status of Ballista.
>
> Ballista is now capable of running some of the TPC-H benchmark queries against the 1TB data set. I documented the benchmark results for DataFusion, Ballista, and Apache Spark for reference, here:
>
> https://github.com/andygrove/ballista-research/wiki/Ballista-benchmarks-results
>
> Here is a chart showing these initial results. As noted on the wiki page, I didn't put too much thought into the configurations used but they are at least documented. I now plan on iterating on my benchmark process to measure memory usage and show scalability as executors/cores are added for each solution.
> Note that I hit failures with all three solutions and this is why there are columns missing from the chart.
>
> I'll try and get something more compelling written up for a blog post to coincide with the upcoming release of Ballista 0.5.0 but I figured folks might be interested in these informal interim results.
>
> Thanks,
>
> Andy.
>
>
>