I benchmarked this a couple weeks ago, and just now got around to
putting the results in a presentable form.
I took some of the more interesting cases from the B&M test Peter
shared upthread (best and worst performers in a quick test) and ran
them with a range of "m" values.
This is exploratory w
On Mon, Nov 25, 2024 at 10:20 PM Peter Geoghegan wrote:
>
> I suggest using a test program for this that Tom wrote nearly 20 years
> ago to validate changes that were made to the Bentley & McIlroy qsort,
> available from here:
>
> https://www.postgresql.org/message-id/18732.1142967...@sss.pgh.pa.u
On Mon, Nov 25, 2024 at 7:14 AM John Naylor wrote:
> To evaluate this technique further, it'll need some work to handle
> duplicates well.
I suggest using a test program for this that Tom wrote nearly 20 years
ago to validate changes that were made to the Bentley & McIlroy qsort,
available from h
Attached is a very rough and limited proof of concept of $subject when
tuplesort uses abbreviated keys. It only works with int64:
Demo:
--setup
drop table if exists test;
create table test (a bigint);
insert into test select (1_000_000_000 * random())::bigint from
generate_series(1,1_000_000,1) i
