PostgreSQL/FreeBSD performance and scalability on a 40-core machine [pdf](kib.kiev.ua)
kib.kiev.ua
PostgreSQL/FreeBSD performance and scalability on a 40-core machine [pdf]
https://kib.kiev.ua/kib/pgsql_perf.pdf
5 comments
The DragonFly benchmark results included in the reference at the end of this document include DragonFly, FreeBSD 10.0, Centos and Debian. In short, the other operating systems they benchmarked don't display the substantial scalability problem observed on FreeBSD in their test.
http://lists.dragonflybsd.org/pipermail/users/attachments/20...
Note that there are some concerns with the benchmark technique used here (both the test that's the subject of this post, and the DragonFly testing). Running the pgbench test tool on the same host that's running the database under test can severely impact the results in ways that don't reflect what you'd see in a real deployment. See this post from Attilio Rao for more detail: http://lists.freebsd.org/pipermail/freebsd-performance/2014-...
The test here was done to push PostgreSQL on FreeBSD to the limit, in order to understand and then eventually fix the bottlenecks; it's not directly useful as an indication of the performance one should expect from a real deployment.
http://lists.dragonflybsd.org/pipermail/users/attachments/20...
Note that there are some concerns with the benchmark technique used here (both the test that's the subject of this post, and the DragonFly testing). Running the pgbench test tool on the same host that's running the database under test can severely impact the results in ways that don't reflect what you'd see in a real deployment. See this post from Attilio Rao for more detail: http://lists.freebsd.org/pipermail/freebsd-performance/2014-...
The test here was done to push PostgreSQL on FreeBSD to the limit, in order to understand and then eventually fix the bottlenecks; it's not directly useful as an indication of the performance one should expect from a real deployment.
Here are the patches referenced in the article:
https://kib.kiev.ua/kib/pig1.patch.txt
https://kib.kiev.ua/kib/patch-2
https://kib.kiev.ua/kib/pig1.patch.txt
https://kib.kiev.ua/kib/patch-2
When did we stop getting [scribd] suffixes for PDF links?
About a week ago. https://news.ycombinator.com/item?id=7928852
It seems your public domain suffix list needs updating (kiev.ua is a public domain). Someone's posted a link to a relevant Mozilla project recently: https://publicsuffix.org/
I feel like this is obligatory: http://xkcd.com/619/
Different strokes for different folks. I will be very happy when the PG >9.3 performance regressions are addressed in the FreeBSD kernel.
s/uses/pays for/, and a joke becomes a somber realization.
* Peak performance came at the number of threads == number of cores (vs hardware threads, being double that).
* How does the same hardware profile when running Linux?