Re: ML-based indexing ("The Case for Learned Index Structures", a paper from Google)
| От | Jonah H. Harris |
|---|---|
| Тема | Re: ML-based indexing ("The Case for Learned Index Structures", a paper from Google) |
| Дата | |
| Msg-id | CADUqk8VYXaA3_OC=pwvR90MwKVp2tWrEVN_o7L5AB57zedxyuA@mail.gmail.com обсуждение исходный текст |
| Ответ на | Re: ML-based indexing ("The Case for Learned Index Structures", a paper from Google) (Peter Geoghegan <pg@bowt.ie>) |
| Ответы |
Re: ML-based indexing ("The Case for Learned Index Structures", a paper from Google)
|
| Список | pgsql-hackers |
On Tue, Apr 20, 2021 at 3:45 PM Peter Geoghegan <pg@bowt.ie> wrote:
On Tue, Apr 20, 2021 at 12:35 PM Chapman Flack <chap@anastigmatix.net> wrote:
> How would showing that to be true for data structure X be different from
> making a case for data structure X?
You don't have to understand the theoretical basis of B-Tree indexes
to see that they work well. In fact, it took at least a decade for
somebody to formalize how the space utilization works with B-Trees
containing random data. Of course theory matters, but the fact is that
B-Trees had been widely used for commercial and scientific
applications that whole time.
Maybe I'll be wrong about learned indexes - who knows? But the burden
of proof is not mine. I prefer to spend my time on things that I am
reasonably confident will work out well ahead of time.
Agreed on all of your takes, Peter. In time, they will probably be more realistic. But, at present, I tend to see the research papers make comparisons between learned vs. traditional pitching the benefits of the former without any of the well-known optimizations of the latter - as if time stood still since the original B-Tree. Similarly, where most academic research starts to fall apart in practicality is the lack of addressing realistic write volumes and related concurrency issues. I'm happy to be disproven on this, though.
Jonah H. Harris
В списке pgsql-hackers по дате отправления: