Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks

[energyman]

it was from 2006, yes. But hey, since then nothing really changed. The state of fs in *BSD still sucks. No changes there. The pattern is pretty simple. Over the years the *BSD fanboys are hyping some stuff (like softupdates, or their pretty new scheduler), just to get blown out of the water as soon, as benchmarks by non-*bsd devs/fanboys are coming in.

Some things never change.

[rivald]

heh heh... Yes, nothing has changed.

If we're going to dig up old junk on the web:

http://people.freebsd.org/~murray/bsd_flier.html

That really brings me back... mounting filesystem async in production. Great idea. Performance is king, after all.

If you're unfortunate enough to make a living at systems/network administration, you'll come to realize how much each OS sucks in its own special way. You'll shout curses at 3:00 in the morning at microsoft, linux developers, sun, freebsd developers, openbsd developers, apple, cisco, netgear, hp, dell, sgi, checkpoint, and the rest of them.

[kraftman]

Nice, a benchmark from 2006. I seem to recall a benchmark from 2000 where windows 2000 blew away a heavily tuned Redhat. Perhaps we can drag that up and make the argument that windows blows linux away? If old benchmarks are okay, maybe we should use them? Perhaps the benchmarks should have included deleting the files, too? Looks like the 2006 benchmarks weren't so favorable to linux.

I think they should rerun it with OpenSolaris, FreeBSD, and Ubuntu, all with debugging turned on.

I don't understand why linux fans try so hard.

I don't understand why you can't see that FreeBSD fan started. I can show you that Linux just outperform your favorable system:

http://new.isc.org/proj/dnsperf/OStest.html

Linux 2.6.20(!) kicked ass your FreeBSD-7-CURRENT 200708...

I'm sure the whole point of the benchmark was to show linux as better than everything else, but really... at least take real releases and not betas.

Bulshit... You didn't see previous one. It's something common that freebsd fanboys usually start flames.

EDIT:

Benchmarks made by bsd devs are REALLY objective XD. I fell from my chair when I saw their last revelation - they said that FreeBSD 7 is 15% faster in MySQL (or PostGreSQL), then I searched for some benchmarks and even Linux 2.6.22 was faster.

@energyman

Exactly.

[kebabbert]

and there are many, many links showing linux scaling very well with lots of CPUs.

Hmm...

SunOS has nothing to do with Solaris. When they changed the name they changed the basis of their kernel too. 'Original' solaris is also known for tons of broken crap. Broken tar, broken find, etc pp.

You could also argue, that Phoronix used the slowest linux out there, instead of something fast.

A quick googling showed some examples of bad scaling:

http://blogs.zdnet.com/BTL/?p=8769
"And if I’m not correct, you don’t currently see the level of geometric performance increases on Linux above 16 cores like you do with UNIX. The maturity in the Linux kernel for this level of enterprise performance and stability on this type of hardware just isn’t there yet."


"Gracenote, which provides a media recognition and metadata service for MP3 users (the CDDB database familiar to iTunes users), agrees. "We found the threading model in Linux was problematic. You get to a certain number of concurrent threads and the OS just slows way down," says Matthew Leeds, vice president of operations at Gracenote. Solaris "just works for us.""




But bad scaling is of course to be expected. Linus Torvalds says that Linux evolves just as nature; evolution. Whenever something is broken in Linux he will redesign it into something better. Just as evolution.

To me that is less optimal. How can something be good if you totally redesign it from scratch all the time? v1.0 is never good. You have to polish it for a long time before it is stable and good. Just like Windows, they rewrite everything and first at SP2 windows becomes quite stable. The first iterations suck badly. Dont you agree? And, the constant redesigning is also one of the reasons Linux is never back compatible. There are also lots of links on this. And, also the reason behind the bad coding and all the errors and bugs in the Linux kernel, that Linux kernel dev Andrew Morton speaks of:
http://lwn.net/Articles/285088/
Now Linux is 6.4 million lines of code. It must be impossible to keep all that monolithic Linux kernel with 6.4 million lines of code, bugg free. How can one KERNEL be that big? If you count the new lines, comments etc, Linux kernel is >10 million lines of code. I remember the whole Windows NT was 10 million lines of code. Bloat? Ridicolous. You know, the less code, the better. Right?

Solaris first revision, SunOS, sucked and at the next attempt SUN knew how to do it right -> Solaris. Linux is at it first revision now, and it sucks when stressed enough, because of all the bugs that Andrew talks of. It is a desktop OS, not a Server OS. Here you see some quick googled links what what happens when you stress Linux. Linux is stable at up to 60% or so utilisation, and Solaris stable up to 100% just as mainframes.

http://www.enterprisestorageforum.co...le.php/3745996

http://lethargy.org/~jesus/archives/...ver-Linux.html

[rivald]

The funny thing is that this benchmark reminds me why ZFS blows away the other filesystems. What the benchmark doesn't show is that ZFS is more reliable.

If you've used ZFS professionally, you'd realize that was also the case... it's better than LVM, Sun's disksuite and the BSD equivalents (probally gvinum on FreeBSD, and raidframe on OpenBSD and NetBSD.)

I'd say ZFS is the real winner here... it's particularly impressive considering how slow Solaris is at certain operations (if you've administered it professionally, you'll know what I mean) - and this is even an RC. Slowlaris no more!

[kraftman]

The funny thing is that this benchmark reminds me why ZFS blows away the other filesystems. What the benchmark doesn't show is that ZFS is more reliable.

If you've used ZFS professionally, you'd realize that was also the case... it's better than LVM, Sun's disksuite and the BSD equivalents (probally gvinum on FreeBSD, and raidframe on OpenBSD and NetBSD.)

I'd say ZFS is the real winner here... it's particularly impressive considering how slow Solaris is at certain operations (if you've administered it professionally, you'll know what I mean) - and this is even an RC. Slowlaris no more!

The same situation can be with Phoronix benchmarks. If something is slower it can be more reliable etc. but it's not a rule.

[rivald]

True. Benchmarks aren't necessarily relevant, anyway. I'm sure there are MS benchmarks out there showing Windows 2008 as being faster than everything else.

For any professional work, I prefer reliable over fast. That's what I like about ZFS (and it's still fast.) I've used AFS in production environments, and it might be the best of all, in some ways, but it's way more of a hassle to manage.

[kraftman]

@kebabbert

You're repeating some myths introduced by some group of people. There is known problem with "bad scalling" on Linux just, because some idiots test it with broken/bad/not Linux friendly library. If you replace it scaling is great.

...You know, the less code, the better. Right?

Total bullshit. What additional features and drivers have to core Linux kernel parts? And on Linux works probably far more people then on other operating systems.

@rivald

ZFS is in many aspects probably the best file system right now.

I'm sure there are MS benchmarks out there showing Windows 2008 as being faster than everything else.

That's for sure

[energyman]

BLABLA

just look at the Top 500 supercomputers. Huge machines with thousands of CPUs.

And now ask yourself 'why are most of them using linux'?
And then ask yourself 'why is nobody using Solaris'?

[energyman]

uh, I was wrong, there is one(!) Supercomputer using Opensolaris (not even solaris... )
http://www.top500.org/stats/list/32/os

Open Solaris 1 0.20 % 18540 20643 2048
compares nicely with the 454 or so Linux using supercomputers...