PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#1

Post by glutamate » Thu, Dec 28, 2017, 2:54 PM UTC

PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE
redbyte.eu

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#2

Post by Tostino » Thu, Dec 28, 2017, 3:33 PM UTC

Oh man, that is some serious SW raid overhead.

Does anyone who knows FreeBSD and Postgres have any idea why it was so much slower for the read only tests?

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#3

Post by brightball » Thu, Dec 28, 2017, 3:34 PM UTC

Wow! The difference in RAID between FreeBSD and the GNU Linux flavors is more than a little bit shocking on the read/write tests. I would not have expected it to be that significant.

I wonder what it will take to fix that in Linux?

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#4

Post by saosebastiao » Thu, Dec 28, 2017, 3:36 PM UTC

I wasn’t expecting such a drastic difference between FreeBSD and the linux distros. I wonder if this is a general BSD thing, or if it was FreeBSD specific (or filesystem or drivers).

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#5

Post by Fiahil » Thu, Dec 28, 2017, 3:40 PM UTC

> 32GiB read only: read only test, the dataset does not fit into the PostgreSQL cache

> 200GiB read only: read only test, the dataset fits into the PostgreSQL cache

How come a larger dataset fit into the PostgreSQL cache and not the smaller one ?

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#6

Post by belthesar » Thu, Dec 28, 2017, 3:44 PM UTC

Fiahil wrote: post #5

> 32GiB read only: read only test, the dataset does not fit into the PostgreSQL cache > 200GiB read only: read only test, the dataset fits into the PostgreSQL cache How come a larger dataset fit into the PostgreSQL cache and not the smaller one ?

I wondered the same thing. Best explanation had to do with the 32 GB dataset being on the system with 32 GB of RAM vs the 200 GB dataset on the system with 256 GB of RAM. Makes sense, but isn’t clear from the article’s perspective.

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#7

Post by gigatexal » Thu, Dec 28, 2017, 3:46 PM UTC

I don’t think it’s linux raid software overhead but on the fly compression decompression of ZFS which could mean the BSDs are doing less IO than their Linux counterparts. I don’t know for sure it’s just a hunch.

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#8

Post by elygre » Thu, Dec 28, 2017, 3:46 PM UTC

Fiahil wrote: post #5

> 32GiB read only: read only test, the dataset does not fit into the PostgreSQL cache > 200GiB read only: read only test, the dataset fits into the PostgreSQL cache How come a larger dataset fit into the PostgreSQL cache and not the smaller one ?

Probably different postgresql.conf-settings. For example, 32GiB instance has "effective_cache_size = 24GB", while 200GiB instance has "effective_cache_size = 144GB"

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#9

Post by kchoudhu » Thu, Dec 28, 2017, 3:49 PM UTC

Wow, this is pretty eye-opening for anyone looking to use FreeBSD in production as a database server.

Anyone from FreeBSD devs have any thoughts on why we're so far behind?

Re: PostgreSQL Benchmark on FreeBSD, CentOS, Ubuntu Debian and OpenSUSE

#10

Post by GalacticDomin8r » Thu, Dec 28, 2017, 3:52 PM UTC

Tostino wrote: post #2

Oh man, that is some serious SW raid overhead. Does anyone who knows FreeBSD and Postgres have any idea why it was so much slower for the read only tests?

It may be related to the multi-threaded page daemon:

https://www.kib.kiev.ua/kib/pgsql_perf_v2.0.pdf