FreeBSD + PostgreSQL = win!

drhowarddrfine

Son of Beastie

Thanks: 870
Messages: 2,720

#2
Only scanning quickly the article, at first I was confused and it looked bad for FreeBSD but the graphs at the end, and the summary, show this:
I was positively surprised by FreeBSD 11.1 which was more than twice as fast as the best performing Linux, despite the fact that FreeBSD used ZFS which is a copy-on-write file system.
 

tingo

Daemon

Thanks: 336
Messages: 1,844

#3
Interested people should also read the HN discussion thread linked to in the blog post. It illustrates why benchmarking is hard, and why talking sensibly about benchmark results is even harder.
 
OP
OP
achix

achix

Well-Known Member

Thanks: 46
Messages: 285

#4
OTOH ZFS might be cool and all (I confess my guilt I have only once tried to fresh install a FreeBSD box with ZFS root and failed, and didn't find time since then) but requires some learning curve, so it would be nice if UFS could match EXT4, unless things were "liberally" configured for linux (e.g. write cache on) and conservatively configured for FreeBSD. The papers unfortunately lacks such details.
 

ronaldlees

Aspiring Daemon

Thanks: 263
Messages: 679

#5
I guess PostgreSQL was always a win for BSD, considering it started at Berkeley, and ended in 1985, although at that time it was just "Postgres" and didn't have SQL incorporated into it yet. SQL functionality was added in 1994. So, maybe the early engines were tailored to a BSD architecture, albeit that doesn't have anything to do with ZFS.
 

Oko

Daemon

Thanks: 763
Messages: 1,620

#6
OTOH ZFS might be cool and all (I confess my guilt I have only once tried to fresh install a FreeBSD box with ZFS root and failed, and didn't find time since then) but requires some learning curve, so it would be nice if UFS could match EXT4
Match in what? RHEL and derivatives use XFS. Ubuntu is using EXT2 for root and is aggressively adopting ZFS. EXT4 until recently could not hold more than 16 TB of data. My ZFS pools these days are routinely close to 100TB
 
OP
OP
achix

achix

Well-Known Member

Thanks: 46
Messages: 285

#7
I guess PostgreSQL was always a win for BSD, considering it started at Berkeley, and ended in 1985, although at that time it was just "Postgres" and didn't have SQL incorporated into it yet. SQL functionality was added in 1994. So, maybe the early engines were tailored to a BSD architecture, albeit that doesn't have anything to do with ZFS.
If that was the case remotely then UFS would outerperform EXT4.
Key ppl in PostgreSQL since 20 years back are diehard Linux boys.
 
OP
OP
achix

achix

Well-Known Member

Thanks: 46
Messages: 285

#8
Match in what? RHEL and derivatives use XFS. Ubuntu is using EXT2 for root and is aggressively adopting ZFS. EXT4 until recently could not hold more than 16 TB of data. My ZFS pools these days are routinely close to 100TB
- Unfortunately it seems EXT4 swept the floor under UFS, specifcally on this benchmark.
- Ubuntu has had EXT4 support since Ubuntu 9.04 (Jaunty Jackalope) (10 years back!!)
- Ubuntu of course uses EXT4 in root :
Code:
achill@ubuntu-achill:~$ mount | grep "/ "
/dev/vda1 on / type ext4 (rw,relatime,errors=remount-ro,stripe=4,data=ordered)
achill@ubuntu-achill:~$
I cannot verify the "until recently could not hold more than 16 TB of data" part, so I believe you in this.
Also you are right about Ubuntu fully supporting ZFS.

How many companies are there in our hood (balkans) with DBs larger than 16TB? And even then there are tablespaces, LVM, a plethora of ways to overcome this. And even then putting everything on the same vol would be an anti-pattern.

In our installation, EXT4 is doing great.
 
Top