Benchmarking ZFS On FreeBSD vs. EXT4 & Btrfs On Linux

[kraftman]

For you information those are the papers:

http://en.wikipedia.org/wiki/List_of...mputer_science

Not sun marketing bull, their lame biased benchmarks and sun paid people bull. Now, troll, show me the papers.

[sbergman27]

Well Steve. First of all, I have studied more math than you ever had, which is evident because you say so ignorant things. Second, I suggest you read a bit more. You know, the hype about ZFS and DTrace, etc - is for real. The Solaris guys ARE good. They KNOW what they are doing. Otherwise noone would cared for ZFS nor DTrace. When the Solaris guys says things, you better listen. They know things. You DONT. They can see trends, they know how much storage the server halls are bying, and they see it grows exponentially. You dont have that information.

When you talk about Moore's law
"Moore's law is about transistor density on silicon. It has nothing to do with disk space."

Let us read more on this law, regarding hard drives part:
http://en.wikipedia.org/wiki/Moore%27s_law
It turns out that hard drives have the same development, called Kryder's law:
"A similar law [as Moore's law] (sometimes called Kryder's Law) has held for hard disk storage cost per unit of information"

And there is a article in "Scientific American" about Kryder's law:
http://www.scientificamerican.com/ar...id=kryders-law
Which says that "says that magnetic disk areal storage density doubles annually"

Do you know enough math to understand what this means? It means EXPONENTIAL GROWTH, the same as Moore's law. And if you dont know theory about asymptotics, let me tell you. Exponential growth is a very bad thing, it grows extremely fast. It actually, grows exponentially. Therefore you are wrong on this. Your premise is false, and your entire reasoning is FAIL. And you say I dont understand arithmetic and math? Jesus. What are those Linux guys out there? Uneducated all of them? I cant help but wonder. Why am I wasting time pointing out errors in their juvenile reasonings?

You're trying to evade the issue. (The issue being whether designing a 128 bit filesystem today is reasonable or completely over-the-top.) I'll respond only to the one relevant bit of your response.

Kryders' Law is demonstrably optimistic. In 1996, a Seagate ST225 20MB drive sold for about $229. (I remember because I bought one.) In the intervening 24 years, if Kryder's estimate were correct, we'd have 335TB drives on the shelves today. We don't. If instead, you assume that capacity has doubled every 18 months, it extrapolates about to about 1.4TB today. Clearly, viewed over the last 24 years, an 18 month doubling period is far more accurate than the "doubling annually" supposition. Then again, Kryder is talking about data per unit area, and drive areas have decreased. The important and relevant bit is that from a commercial standpoint, the cost per dollar has halved only about every 18 months. That, and the practical matter of "how many physical drives does it take to achieve X capacity", and not data per unit area, are the relevant points regarding.

But even if we assume that filesystem size requirements doubled annually, we're still looking at 32 years to grow from what 32 bit can support to what 64 bit can support. And another 64 years to go from what 64 bit can support to what 128 bit can support, for a total of 96 years. If we (*very* generously) assume that some apps today require 48 bits, (and can anyone cite an example of that?) that's still 80 years using your overly optimistic growth estimate. I ask again, do you really think people will be mkfs'ing new filesystems with ZFS in 80 years?!

In a nutshell... and just so we don't go too far off-track... my position is that ZFS is a fine filesystem for *Solaris, but wouldn't be a good fit for Linux even aside from the licensing issue. btrfs *is* a good fit for Linux. btrfs development is proceeding very well. And 128 bitness in a filesystem is just plain silly in 2010.

[kebabbert]

You're trying to evade the issue. (The issue being whether designing a 128 bit filesystem today is reasonable or completely over-the-top.) I'll respond only to the one relevant bit of your response.

Kryders' Law is demonstrably optimistic. In 1996, a Seagate ST225 20MB drive sold for about $229. (I remember because I bought one.) In the intervening 24 years, if Kryder's estimate were correct, we'd have 335TB drives on the shelves today. We don't. If instead, you assume that capacity has doubled every 18 months, it extrapolates about to about 1.4TB today. Clearly, viewed over the last 24 years, an 18 month doubling period is far more accurate than the "doubling annually" supposition. Then again, Kryder is talking about data per unit area, and drive areas have decreased. The important and relevant bit is that from a commercial standpoint, the cost per dollar has halved only about every 18 months. That, and the practical matter of "how many physical drives does it take to achieve X capacity", and not data per unit area, are the relevant points regarding.

But even if we assume that filesystem size requirements doubled annually, we're still looking at 32 years to grow from what 32 bit can support to what 64 bit can support. And another 64 years to go from what 64 bit can support to what 128 bit can support, for a total of 96 years. If we (*very* generously) assume that some apps today require 48 bits, (and can anyone cite an example of that?) that's still 80 years using your overly optimistic growth estimate. I ask again, do you really think people will be mkfs'ing new filesystems with ZFS in 80 years?!

In a nutshell... and just so we don't go too far off-track... my position is that ZFS is a fine filesystem for *Solaris, but wouldn't be a good fit for Linux even aside from the licensing issue. btrfs *is* a good fit for Linux. btrfs development is proceeding very well. And 128 bitness in a filesystem is just plain silly in 2010.

Jesus. There are so many errors in your posts, I dont know where to begin. But let me point out some errors in your logic, because you apparently dont see the errors. I, the "person that dont know math", will lecture you on math and formulas. I will give the big picture and not the details. I dont want to spend hours and hours pointing out all errors in your reasoning.




First of all, you talk about me avoiding the issue. No, I dont. There is no issue as you have presented it. Because your presentation is wrong. You could as well say "the moon is made of cheese" - and when I dont discuss that false statement you say that I evade the issue??? No I dont, the moon is not made of cheese. There is nothing to discuss about the moon. Regarding filesystems, no, there is nothing to discuss because your premise is wrong. But I can try to explain again why you are wrong. Here goes.



Second, you talk of Kryder's law being overly optimistic, by giving some calculations to disprove his law. You have missed the point. You dont understand it. Let me tell you the point. First of all, it is not an exact mathematical formula which you have "disproved by giving some calculations". It is only an empiric observation, not a formula. You can use it the way you do. Wrong.



Third, the important point with Kryder's law is: it points out that the storage growth is exponential. That is the only important thing to remember. That Kryder's law says: c^n where c is some constant > 1. The constant is not important, it does not matter if it is 1.4 or 1.9 or whatever. And if you have studied asymptotics and Big Oh, then you know what conclusion to draw. If you have not, study it. I can not write a crash course here. The point is, as the growth is exponential, we can expect huge increases in short time (this is another point you dont understand). HUGE increases. Let me ask you, how huge? You say: big. No, you dont understand. It is more than big. Exponential is HUGE. It is hard to describe informally. But study asymptotics and you will understand. We are only interested in asymptotics, a ball park. You are wrong.



Fourth, you talk about 96 years, or so, until we need ZFS. Now that is wrong because of several reasons. You would be correct, if we started today at year 1. But we dont start today at year 1. Today, we are already at year 60(?) or so. So there will not be another 96 years before we need ZFS, there will only be 36 years (if your numbers where correct). Wrong.



Fifth, Actually, today we have far surpassed 32 bit storage. 32 bit storage is only 2TB. We are storing much more than 2TB today. We are already in the realm of PetaBytes. CERN stores today 8 Petabytes. The first year CERN runs LHC, they will need to store double that data, 16 Petabytes. I dont know how much Google store, but indexing the whole internet needs much data storage.

So, how much is a 64 bit filesystem capable of storing? It is just some millions Petabytes! How long does it take to increase something a million times? Well, how long time took it to go from 4KB RAM to 4GB RAM (a million times increase)? It took only, something like 15(?) years or so. This is because the growth is exponential, the development is HUGE. Faster than you can imagine. So, if we store exponentially more information all the time, then it will only take maybe 15 years before we need to store millions of Petabyte, today we store only 10s of Petabyte. If the exponential constant is lower for Kryder's law than for Moore's law, it may take 20 years instead. In just a couple of years from now, CERN needs to store 100 Petabyte.

So in a short amount of time, humanity need to store more than 64 bits, we need a 65th bit. And filesystems live for a very long time. In Enterprise halls, they live for decades. Preferably nothing is changed in Enterprise halls, they run for decades. Actually, what I say here, is basically the same as the ZFS developers said in my link, read it again. How can you accept my explanation, but not the ZFS developer's? Pure bias? So you are wrong. We very soon need more than 64 bits.

Should we settle for a 70 bit filesystem then? And then in a few years we need a 80 bit filesystem? etc. Why not 128 bit at once, and then all worries are gone for ever? Humanity will never need more than 128 bit filesystems. So your calculations are wrong. And the ZFS developers calculations are correct. They dont bull shit, or talk marketing. CERN, among others, really WILL need a 65th bit in some 15 years from now. It IS true, it WILL happen. If you object, you are free to point out my errors.


Sixth, there are more errors, so I just say: Wrong.


Seventh, Wrong.


Next time, before you claim someone dont know arithmetic, I suggest you checkup your own arguments first. Maybe he has a PhD in math?

In a nutshell... and just so we don't go too far off-track... my position is that ZFS is a fine filesystem for *Solaris, but wouldn't be a good fit for Linux even aside from the licensing issue. btrfs *is* a good fit for Linux. btrfs development is proceeding very well. And 128 bitness in a filesystem is just plain silly in 2010.

I dont say BTRFS is a bad filesystem. I think that if all the high goals are achieved, then it will be a good ZFS copy. And even if BTRFS does not succeed in giving data integrity, it will hopefully give more data integrity than the rest: raid5, raid6, XFS, JFS, etc. I also agree it is a good fit for Linux. I suspect BTRFS will be the best common filesystem out there (except ZFS of course)

But I dont agree that ZFS would not be a good fit for Linux. FreeBSD people likes ZFS and uses it a lot, so why can not Linux people use ZFS? Especially as Linux community for long has wanted ZFS. I mean, when you upgrade your Linux system and it breaks (because of unstable ABI) you typically need to reinstall to get working Linux back. If Linux would use ZFS instead, you would just do a snapshot before upgrading, and then reboot into the earlier snapshot via GRUB if the upgrade breaks your install. Then you delete the upgrade and then your system is back where it was before the upgrade. How can this not be a good for Linux? And also, I suspect Linux people's data is also valuable and need data safety? How can this not be good for Linux?

"128 bitness in a filesystem is just plain silly in 2010" - Wrong. I hope I taught you something today.

[kebabbert]

Prove what wrong. I called you crazy. Your going on about what people think and whats common in a relationship? Crazy you are.

Fair enough. You are free to think whatever you want of me. But, it does not matter, as long as ZFS is better than BTRFS.

Lets get inside the head of oracle for a second. They now control ZFS and BTRFS. Why in the hell would they spend a bucketload of money developing BTRFS to be worse then ZFS? They wouldnt. Would you start saving your money to buy an old pentium 2 desktop as a replacement to your current setup? this is unlikely. The thought that oracle wants to waste money in this way is inconceivable.

Or, let us look it this way instead.

Sun has invested many years or research and much money and lots of hard work into develop ZFS to be the best common filesystem in the world. Sun earned money on ZFS storage servers, they where profitable and sold very much. ZFS has a proven track record. It delivers.

Oracle has not invested much in BTRFS, one full time developer has been paid for some time. BTRFS has not delivered yet, it is only hype at this stage.

Oracle earns money now, on ZFS storage servers and on Solaris 10 that runs ZFS. Is this a bad thing, does Oracle want to change this situation? Why would BTRFS development surpass ZFS? If so, ZFS storage servers would not sell anymore. Would not it be better if Oracle killed BTRFS, or made it bad? Then everyone would need to buy Solaris 10 and ZFS storage servers to get the only filesystem that gives data protection. Or, should Oracle invest lots of money to make a copy of a filesystem and give it away for free?

What do you think?

I dont care what papers you provide, or what links you throw at me because you are really not worth my time.

At least I am not lying or making things up. My links are credible and some people trust the links. It is not just air or false rumours spread by competitors or FUD.

I have found that dispite being only 1 man who needs to eat and sleep, Michael does a damn good job benchmarking. From what i see in the charts, BTRFS kicks ZFS ass in a fucking development stage.

I dont argue about this one actually. I dont care if BTRFS is faster than ZFS on single disc. Who cares? With ZFS, your data is safe, that is the important bit.

When I talked about ZFS speed in Enterprise servers, I just wanted to show to Energyman, mr Muhahahaha, that ZFS does provide plenty of speed if need be. That he is wrong laughing like that. That I should be the one laughing at him.

But regarding the speed thing, I dont care. Actually. I just wanted to show that ZFS gives extreme performance, if the need arises. But the only thing that is important, is data safety. That, I do care about.

If you would like to argue that ZFS keeps your data safer, go ahead, but your competing with a filesystem still in development. If you for one second belive that a development FS is going to be a safer bet than a competed one then you have problems beyond repair.

I dont compare a file system under development with a released one. I am only saying that I doubt BTRFS will be able to provide data safety, as no one has succeeded during 30(?) years. Raid5 does not. XFS does not. ReiserFS does not. etc. Facit, the research from the real world shows it is very difficult to provide data safety. I doubt BTRFS will succeed on this one. It requires a huge experience from Enterprise storage, which this single developer probably does not have. In reality, I expect BTRFS to be as bad as XFS, JFS, etc with respect to data integrity.

For the other parts, speed and performance, I dont care. I know ZFS wins it easily if there is a need. Just go and fetch a large Enterprise ZFS machine, and the match is over.

But as i said, it is unlikely oracle is making a 2nd rate FS, especially when they have all the ZFS licencing cards.

Or, it will be like I say: Oracle will stop investing in BTRFS now when Oracle already has the best product on the common market. It does not make sense to invest in two similar products, targeting the same customers. If so, you always kill the lesser product. In this case, maybe Oracle will continue to pay this single developer and not allocate more resources to BTRFS - just to make the Linux people happy. Seriously, if Oracle really where serious with BTRFS, then Oracle would have assigned whole team of developers to BTRFS. Preferably some ZFS developers would have been reassigned to BTRFS. Dont you understand this? Actually, I suspect BTRFS will never lift. There is no resources allocated to BTRFS. Not a huge team. ZFS is already profitable. And best.

You dont see this? And you call me crazy? Maybe you just dont understand things?

kraftman is right, fucking troll.

Yes, I agree except the "fucking" part. Kraftman is really a troll. But he is not a "fucking troll", only an ordinary troll. We like Kraftman too much to say he is a "fucking troll"!

[kebabbert]

@kebabbert:
I just don't like holy wars. And i don't like off-topics and spam.

All that you said on this forum was pure off-topic and spam. Article says that this tests are focused on DESKTOP using, so all that you said about enterprise is OFF-TOPIC! And as you repeated your off-topic posts many many times - you are doing nothing more than spaming.

Look, i really don't care about enterprise. I don't care is btrfs a killer of ZFS. I don't care is ZFS always a winner. I don't have matrix of 100 discs to test it.

Brtfs own ZFS's ass on phoronix DESKTOP tests, so what? It really means nothing to normal desktop user because normal user won't switch to another OS only because of it's filesystem! Isn't that logical and clear?

Phoronix is NOT the place to compare enteprise things!!! Even if your arguments are true - they mean NOTHING on this forum.
When i was clicking on comments button on this article i wasn't expecting anybody could even think to write about enterprise things, simply because it's totally OFF-TOPIC and out of phoronix's horizon!

And look at your posts here:
http://www.phoronix.com/forums/showt...t=24728&page=5
Maaan... they are horrible, that level of logic and drawing of conclusions should be denied, should be banned. Really. Im not joking now. Looks like you were sleeping on those logic lessons, or maybe your teacher was sleeping, dunno...
Also you are just "copy-pasting" your posts from one thread to another,
and that's really not nice.

One more thing - even if you were right about enterprise i wouldn't believe you because your posts really look like a fanboy's nightmare.

Please believe me that i'm not gnu/linux fanboy, I am using many many systems, and i simply can NOT tell that one, concrete system is is the best.

I don't want to insult you, I'm only saying that you really should to think about what are you doing here.

Sooo... can we end this useless OFF-TOPIC war?

The thing is, if you go back into history you will see that Kraftman attacked me several times. For instance, I once posted something casual, and Kraftman immediately attacked me "famous troll is back" out of nowhere. Where where you when Kraftman attacked me? I dont see how you can attack me when I defend myself? Are you always like that? Attack the defender, and support the attacker?

When I talked about weak points in Linux, Kraftman said I was FUDing. I was not. I just posted benchmarks, etc. I did not lie. And then it escalated between Kraftman and me. He can not stand Linux criticizm and attacks Linux criticizers. Then I defend. Then it escalates. Instead, he should post counter links that disprove my research papers, benchmarks, white papers, etc.



And when you say to me: "that level of logic and drawing of conclusions should be denied" - then prove that I am wrong. Go ahead. No, you can not? Well, if you can not point out errors in my posts, then please dont say so.

I understand you get upset when I criticize Linux, by posting research papers, white papers, official benchmarks, interviews with Linux developers, etc - that is ok. You have the right to be upset when you see how bad Linux is in different aspects. I understand you want me to be silenced. That you want all Linux criticizers to be silenced. That is ok with me. But the West does not work like that. All criticizers can say what they want, as long as they dont lie or FUD. Relevant criticizm is always important to the discussion, or dont you agree?

On the other hand, to me, I welcome negative criticism of Solaris if there are any. I want to read about it. Please post links about bad aspects on Solaris, if you have any. I want to learn more on Solaris. Also bad things. I mean it.

But. I. Do. Not. FUD nor Lie. Ok?

[kebabbert]

For you information those are the papers:

http://en.wikipedia.org/wiki/List_of...mputer_science

Not sun marketing bull, their lame biased benchmarks and sun paid people bull. Now, troll, show me the papers.

I didnt understand this. What do you mean? Do you mean you want to see papers from those journals? Or what do you mean? Could you please be more explicit, it is hard to understand you. (How many times have I said that? 30 times now?)

[kebabbert]

You just FUD and you should be banned. You should also stop posting FUD. I don't care if someone said you something, why should I?

There are several people that wants you banned. I wonder what the moderator say when I say that? What do you think, kratfman?

Those aren't papers and some (sap ones) aren't backing up what you claim. Cut the crap talk FUDer and troll.

Lets get back to my question. What is FUD to you? A linux criticizer, is he FUDing? According to wikipedia, a FUDer is someone who lies and makes things up. Why do you imply I lie? I do not. I only quote others.

Solaris is dead. Prove me wrong moron:

http://www.linux-mag.com/id/7749/

This is much better, kraftman. You post links! However, you should preferably post credible links. Not post links to a comment in a blog. Preferably, you should post links to researchers, professors, senior experts, solaris developers (if you want to prove something bad about solaris, it is not good to post to linux developer when you want to prove a bad thing about solaris - then you should post instead official benchmarks which proves your point), etc. If you can not, then post to market people, linux magazines, etc. If you have studied at university then you would know which sources are better than other sources.

I did not understand what you want me to prove. Could you be more clearer? "Solaris is dead"? I mean, it is sold today. People runs it today. Development continues today. etc. I dont understand what you want me to prove?

[kebabbert]

Great! Kraftman, thanks for your link http://www.linux-mag.com/id/7749/

Your link supports me against this "smitty" person. He FUDs and lies a lot it seems.

False, false, false.

I wrote earlier:

Even Linus T considered to change license to get access to Solaris tech - why would he do that if Solaris tech is only slightly better? Linus said something like "if Solaris is released as GPL v3.0 then I may change license of Linux to GPL v3.0 too".

To that, "smitty" answered:

More misleading quotes. What a surprise, coming from you. /sarcasm

Linus was thinking of possible reasons he might move on to GPLv3, and the only thing he could possibly think of was getting access to Solaris code. He then went on to conclude that it wasn't worth it and that Linux really had nothing to learn from Solaris.

But your link, Kraftman, shows that "smitty" is wrong and probably lies and FUD. Ok, so now we know "smitty" probably also is a FUDer and liar.

Your link, Kraftman, points further to this link:
http://lwn.net/Articles/237905/
Where Linux T discusses about Linux and Solaris and GPL v3.0. Smitty's quote above is not true. Linus T does not say that "Linux has nothing to learn from Solaris" or "it is not worth to license Linux under GPL v3.0 just to get access to Solaris code".

Linus T says in the link above:
"And yes, maybe ZFS is worthwhile enough that I'm willing to go to the effort of trying to relicense the kernel. But quite frankly, I can almost guarantee that Sun won't release ZFS under the GPLv3 even if they release other parts. Because if they did, they'd lose the patent protection.

And yes, I'm cynical, and yes, I hope I'm wrong. And if I'm wrong, I'll very happily retract anything cynical I said about Sun. They _have_ done great things"


smitty, are you a liar and FUDer? Linus T does not say the things you claim. You just lied and FUDed, do you agree?

[sbergman27]

Next time, before you claim someone dont know arithmetic...

So we come to the real crux of the "discussion". I hurt your feelings. I wish I could honestly say I was sorry about that.

At any rate, now that you've firmly established that all 64 bit filesystems are lacking (thank you), I should point out that despite the marketing rhetoric, ZFS is a 64 bit filesystem. Walked right into that one, didn't you? :-)

-Steve

[wake_up]

Can we all stop this?
Because right now this threat is looking like never-ending story...

What do we want to show exactly? Because right now it's looking like a war of fanboys...

Is it worth it? Are we all working in enterprise's huge halls?

You really believe that any sysadmin of huge system is reading phoronix forum and take whatever you are writing here as irrefutable true?
Maybe sysadmin should throw away all that s/he did before and install other system, because you believe that some fs is better?
I really doubt that would ever happen!
I believe he would rather stay with he's own tested solutions.

The truth is that only time can tell which solution is the best.

@kraftman: you really believe that ZFS sux and is dead? Why? Because Linux don't have it? Don't be such ignorant.

But the only thing that is important, is data safety. That, I do care about.

I also do care about data safety. The fact is that for 12 years of using many OSes and many FSes i didn't loose any data.
And I do care about performance of my raid 0 - it's really important for me mostly because virtual systems/networks i am using, and i think i will soon switch from ext4 to btrfs on my linux system.
So i would like to thank phoronix for btrfs tests!

But regarding the speed thing, I dont care.

I don't care that you don't care. The article was about showing performance of FSes, NOT data integrity.

Especially as Linux community for long has wanted ZFS. I mean, when you upgrade your Linux system and it breaks (because of unstable ABI) you typically need to reinstall to get working Linux back. If Linux would use ZFS instead, you would just do a snapshot before upgrading, and then reboot into the earlier snapshot via GRUB if the upgrade breaks your install.

What? I'ts very easy to fix Linux without reinstalling it, ie. by downgrading package that cause error. You could also have a snapshot of working Linux, it's also not hard work.
Sure it's easier and faster with ZFS, and i never said it's not.
I just wanna say that _I_ can live on Linux without ZFS or DTrace very well, it's not something that _i_ need in Linux.

Right now i'm mainly using Linux, I was using FreeBSD for long time and I like this system, and it's very good that they are implementing ZFS into FreeBSD. And i really don't like what Oracle did with OpenSolaris - they killed it for most users. Right now it is only a testing system for features that would be implemented in Solaris... And that sux.

The thing is, if you go back into history you will see that Kraftman attacked me several times. For instance, I once posted something casual, and Kraftman immediately attacked me "famous troll is back" out of nowhere. Where where you when Kraftman attacked me? I dont see how you can attack me when I defend myself? Are you always like that? Attack the defender, and support the attacker?

Then forgive me my ignorance because i red only few of last pages, and from that point of view you looked like attacker, forgive me.