Benchmarks: PostgreSQL 10 Performance Is Looking Good

[phoronix]

Phoronix: Benchmarks: PostgreSQL 10 Performance Is Looking Good

With yesterday's release of the PostgreSQL 10 beta, a number of Phoronix readers and some of my supporters requested benchmarks of this major update to this widely-used SQL database server. Well, for those curious about this early test release, here are some benchmarks...

http://www.phoronix.com/scan.php?page=news_item&px=PostgreSQL-10-Beta-Benchmarks

[andrei_me]

Some of the big changes for this release is more related to select queries, like analytical queries.

In 9.6 the parallel queries were only used on sequential scan, so they became parallel sequential scan

E.g: select count(*) from table


Now with have indexed scan, bitmap scan and other kinds of queries that can be parallized

Eg:
select count(*) from table a
where exists (
select 1 from table b
where a.id = b.id
);

select count(*) from table a
inner join table b on a.id = b.id


And so on

[mark_]

That benchmark is pretty worthless. Only measuring reading and a bit read/write and comparing it with the previous version does not create much valuable information. What about benchmarking a million inserts (like in a production facility), joins on many huge tables (like in gathering statistics over 20 years of data), load from many users (lets say 100), master-standby log shipping performance under these conditions, including the setup time, computing performance of PL/pgSQL functions and comparing it with mariadb, Oracle, DB2? Also combined with different filesystems underneath? Now that is a benchmark.

[boxie]

I would love to test out their clustered mode stuff - that'd be interesting!

[boxie]

That benchmark is pretty worthless. Only measuring reading and a bit read/write and comparing it with the previous version does not create much valuable information. What about benchmarking a million inserts (like in a production facility), joins on many huge tables (like in gathering statistics over 20 years of data), load from many users (lets say 100), master-standby log shipping performance under these conditions, including the setup time, computing performance of PL/pgSQL functions and comparing it with mariadb, Oracle, DB2? Also combined with different filesystems underneath? Now that is a benchmark.

Got a dataset in mind?

[mark_]

ah and (lol?) cute little computer for that benchmark. How about using something like a HPE ProLiant DL380 Gen9 or so?

[mark_]

Got a dataset in mind?

yes. Unfortunately that one cannot be given to the public. But that is a real scenario, while the current benchmark looks like a scenario that covers hosting a private little webforum with 10000 entries.
Some database systems have reference databases. I'd say pick one of that or a comparable one with 50 tables or so, and treat it like being used in production of thousands of products each day over 20 years.

[mark_]

https://wiki.postgresql.org/wiki/Sample_Databases lists some interesting databases that could be benchmarked, like the mouse genome. Create the db structure and pump in GBs of data, then query that stuff. I don't know how portable or complex they are but it sounds like a start.

[uid313]

Have anyone here used the JSON functionality in Postgres?
Is it useful? Is it awesome? Do you love it? Or is it cumbersome, and boring?

[Michael]

ah and (lol?) cute little computer for that benchmark. How about using something like a HPE ProLiant DL380 Gen9 or so?

If only I had one of those....