Horton

Re: Polyphase merge is obsolete

Peter Geoghegan <pg@heroku.com>

View thread View on postgr.es
From: Peter Geoghegan <pg@heroku.com>
To: Heikki Linnakangas <hlinnaka@iki.fi>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2016-10-14T23:30:58Z
Lists: pgsql-hackers 
On Wed, Oct 12, 2016 at 10:16 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
> Let's switch over to a simple k-way balanced merge. Because it's simpler. If
> you're severely limited on memory, like when sorting 1GB of data with
> work_mem='1MB' or less, it's also slightly faster. I'm not too excited about
> the performance aspect, because in the typical case of a single-pass merge,
> there's no difference. But it would be worth changing on simplicity grounds,
> since we're mucking around in tuplesort.c anyway.

This analysis seems sound. I suppose we might as well simplify things
while we're at it.


-- 
Peter Geoghegan

Commits

  1. Fix and clarify function comment on LogicalTapeSetCreate.

  2. Refactor LogicalTapeSet/LogicalTape interface.

  3. Replace polyphase merge algorithm with a simple balanced k-way merge.

  4. logtape.c: do not preallocate for tapes when sorting