Benchmarking ZFS On FreeBSD vs. EXT4 & Btrfs On Linux

[kraftman]

Kebabbert wrote:
"Because you have not read this thread from the beginning, I post this article again. Here is a link to a research paper about how XFS, JFS, ReiserFS, etc does not protect your data - which is what I have claimed earler."

http://www.zdnet.com/blog/storage/ho...ta-at-risk/169

No, wrong again Kraftman. If you read my text, I actually say: "on that blog there is a link to RESEARCH PAPER". So I mean that you should go to that blog, to see the actual research papers. Of course I could have posted the link directly to the research paper, would that have been better?

Great! You have shown at least one paper which is important! I'm waiting for others. This research was done in 2006 and let's take a look:

from the paper:

Second, to remedy the problem of failure handling diffusion, we develop
amodified
ext3 that unifies all failure handling in a Centralized Failure Handler (CFH).
We then showcase the power of centralized failure handling in ext3c, a modified
IRON version of ext3 that uses CFH by demonstrating its support for flexible,
consistent,
and fine-grained policies. By carefully separating policy from mechanism,
ext3c demonstrates how a file system can provide a thorough, comprehensive, and
easily understandable failure-handling policy.

and the response from the lkml:

Thanks, I think most of the ext4 developers have already read these papers.
They are definitely of interest, and reducing e2fsck time is a very important
area of development these days.

We have already implemented the IRON FS journal checksumming feature, and
this will go into ext4 shortly. There is lots of work needed in order to
implement checksumming for the rest of the filesystem, so if you are
interested to work on this please let us know.

So, it seems this is no longer the case when comes to Linux file systems (at least when comes to Ext4 and we can be nearly sure the same about btrfs). It's very probably Ext3 also got those patches, because they were done like stated in the paper, so it was probably a matter of merging them.

Of course it is not credible to post to blogs. But if the blog links to research papers then you can post to that blog. Or if the blog links to for instance, official benchmarks. It is not necessary to link directly to the research paper or the official benchmark.

True.

Jesus Kraftman, we had this discussion before. Earlier I posted to a Sun blog with links to official SAP benchmarks (Solaris wins over Linux) on www.sap.com,

It doesn't proof Solaris wins over Linux, but only some server which runs Solaris wins over server which runs Linux.

but you said something like: the SAP benchmarks where FUD and created by Sun. You said that SAP have partnership with Sun and therefore SAP favours Solaris in official benchmarks. I showed you links that SAP have partnership with several Linux companies - so why did SAP not favour all Linux companies? Jesus Kraftman.

I showed you links some where SAP doesn't like and attacks Linux, because it's not 'American way', so I thought some results can be biased. However, I don't consider now those benchmarks where invalid, but they're not proving what you're claiming.

When Linux wins, everything is good. It does not matter if Linux uses gcc v4.xx 64bit vs OpenSolaris uses different gcc v3.xx 32 bit - you think that are good benchmarks and dont complain. But when Solaris wins, that is bad offical benchmark published by SAP and you say companies favour Sun and it is FUD benchmark. Unbelievable logic.

It would be the same logic as yours in this case if this were true what you're saying here. It does matter for me if there are different GCC used etc.

Jesus Kraftman. If you had read my posts, then you would have seen those research papers. I do not FUD. Read this post again if you want to see some research papers. I have more research papers about data corruption that I can post if anyone asks me.
http://www.phoronix.com/forums/showp...1&postcount=44

As far if I remember you have shown only one research paper related to what you're claiming, but it seems it's obsolete now. Waiting for other papers which backup your claims.

Fine. I understand I FUDs a lot, in your reality. But, you on the other hand have never linked to any research papers or almost no credible links at all. And you have CONFESSED you do FUD. So, really, I dont really understand how you can accuse me of FUDing? (But, there are lots of things I dont understand about you). So it is ok if you think I FUD, especially as you can not prove that I FUD or lie. It is just more FUD from you, about me.

I can probably only suspect you're lying (I showed you definition of 'lie'). I'm thinking nearly the same about you, so who's right?

But I can prove that you FUD, I just link to the post where you confess you are a FUDer. Done. Kraftman = FUD and lies.

I think many people FUD, intentionally or not. If someone FUDs intentionaly like you, then such person is a FUDer and a troll too. I consider Kebabbert = FUD, trol and lies.

[smitty3268]

I did not understand this one. I explained that, for various reasons, top500 does not say much about a technique or OS that is used. PowerPC at 750MHz is the 5th best CPU in the world you think? No? Good.

HPC is very different workload than enterprise, I'll grant you that. But why only 1 is important and the other means nothing doesn't make any sense to me. It's just your personal opinion.

And anyway, my whole point was that you complained you were repeating yourself and others were ignoring it, which is exactly what you've done. Pot, meet kettle.

It is much easier to do one thing that is specializing than to do a general purpose thing. Study some computer science and you will see.

I have. And in many cases it's true that specializing can simplify a lot of stuff. However, that's not what those HPC supercomputers are. You're setting up a straw man argument here.

The engineering necessary to create and efficiently use tens of thousands of nodes is no simple matter. The fact that you think it is means you know nothing about the subject.

For instance, which is most difficult to reach high performance: a general purpose CPU or a specialized simple GPU?

Judging by the complexity of the graphics libraries and how many transistors are on GPUs compared to CPUs, I'd say the former. Which is mostly because the specialized thing they're trying to accomplish is really frickin tough, in comparison to what a general cpu is usually asked to do. That's the whole reason it requires dedicated hardware, of course.

To do one thing well, is easier to do than to excel at many different things. Common sense. Or, study some more.

That depends entirely on what those things are. Is it easier to send a man to the moon (1 very specialized task) or to put 100 different types of planes into the air? That's a much more general task, but also a much easier one to achieve.

Or perhaps we should switch to getting a date with a movie star versus getting multiple dates with many women. The more general task is much more achievable, even though it's also more general. Simply because the specialized task is harder from the beginning, rather than a simple subset like you are assuming.

Keep setting up straw men, and i'll keep knocking them down.

[kebabbert]

Great! You have shown at least one paper which is important! I'm waiting for others.

That paper was there all the time. I dont know how many times I wrote that, but you said I FUD and lie and never posted any research papers? The paper was there all the time. Maybe I should say this again: The paper was there all the time.

THE PAPER WAS THERE ALL THE TIME.

Kraftman, do you have problems reading text? Or any other kind of problems? I am just asking, I wonder why it is so difficult to communicate with you. It must depend upon something. Problems of some kind. Unless it is pure stupidity?

This research was done in 2006 and let's take a look:

from the paper:

and the response from the lkml:

So, it seems this is no longer the case when comes to Linux file systems (at least when comes to Ext4 and we can be nearly sure the same about btrfs). It's very probably Ext3 also got those patches, because they were done like stated in the paper, so it was probably a matter of merging them.

Interesting. Do you have links too?

I showed you links some where SAP doesn't like and attacks Linux, because it's not 'American way', so I thought some results can be biased. However, I don't consider now those benchmarks where invalid,

Now you consider them not invalid? Great. Sigh. So in a couple of years, you will consider all my posts as valid? But it will take a couple of years, lecturing you on research, on argumentation, on logic, etc?




Kebabbert wrote:
"Jesus Kraftman. If you had read my posts, then you would have seen those research papers. I do not FUD. Read this post again if you want to see some research papers. I have more research papers about data corruption that I can post if anyone asks me."

http://www.phoronix.com/forums/showpost.php?p=139501&postcount=44

As far if I remember you have shown only one research paper related to what you're claiming, but it seems it's obsolete now. Waiting for other papers which backup your claims

Jesus Kraftman. Again: I have posted several research papers on data corruption on ext4, XFS, JFS, raid5, etc. I havtae written this many times. Just read that post, if you want to see those research papers.

-Here it is! I have it here!
-No you dont have it.
-Yes, I have. It is here! Look!
-No, you dont.
-But feel it, ke it, look here. It is here
-No, you dont have it. When will you show me?
-But it is here! LOOK! IT IS HERE!
-No. When will you show me?

Kraftman. You are too much. I feel pity for your teachers. Seriously.

I think many people FUD, intentionally or not. If someone FUDs intentionaly like you, then such person is a FUDer and a troll too. I consider Kebabbert = FUD, trol and lies.

Ok, I could be like this: either you are under 20 years. Or you dont read my posts. If I explain something, and you dont read my explanation, then you believe I am wrong. And you ask the same question again. In either way, you look like quite idiotic.

Clearly, you dont read my posts. That is proven now. Why dont you do that? So, I just wonder: how old are you Kraftman? Seriously. Could answer to this? How old are you?

[kraftman]

That paper was there all the time. I dont know how many times I wrote that, but you said I FUD and lie and never posted any research papers? The paper was there all the time. Maybe I should say this again: The paper was there all the time.

THE PAPER WAS THERE ALL THE TIME.

It is obsolete since some time thus you were FUDing.

Kraftman, do you have problems reading text? Or any other kind of problems? I am just asking, I wonder why it is so difficult to communicate with you. It must depend upon something. Problems of some kind. Unless it is pure stupidity?

The way you're thinking is a pure stupidity. FUDer.

Interesting. Do you have links too?

Of course:

http://www.mail-archive.com/[email protected]/msg02286.html

Now you consider them not invalid? Great. Sigh. So in a couple of years, you will consider all my posts as valid? But it will take a couple of years, lecturing you on research, on argumentation, on logic, etc?

Nope, I doubt I will consider your posts are valid. I just have no objections when comes to those two SAP benchmarks where different servers were tested.

Kebabbert wrote:
"Jesus Kraftman. If you had read my posts, then you would have seen those research papers. I do not FUD. Read this post again if you want to see some research papers. I have more research papers about data corruption that I can post if anyone asks me."

http://www.phoronix.com/forums/showpost.php?p=139501&postcount=44

Jesus Kraftman. Again: I have posted several research papers on data corruption on ext4, XFS, JFS, raid5, etc. I havtae written this many times. Just read that post, if you want to see those research papers.

I don't see papers out there. Can you give links directly to the papers not to some blogs?

-Here it is! I have it here!
-No you dont have it.
-Yes, I have. It is here! Look!
-No, you dont.
-But feel it, ke it, look here. It is here
-No, you dont have it. When will you show me?
-But it is here! LOOK! IT IS HERE!
-No. When will you show me?

Again, where are the papers backing up your claims? As far I can see only one paper, but which is obsolete when comes to some Linux file systems.

Kraftman. You are too much. I feel pity for your teachers. Seriously.

It's not common when I've got to speak with idiots, so adaptation takes some time.

Ok, I could be like this: either you are under 20 years. Or you dont read my posts. If I explain something, and you dont read my explanation, then you believe I am wrong. And you ask the same question again. In either way, you look like quite idiotic.

I want papers, otherwise why should I read the same bull all the time?

Clearly, you dont read my posts. That is proven now. Why dont you do that? So, I just wonder: how old are you Kraftman? Seriously. Could answer to this? How old are you?

I repeat all the time. Show the papers which backup your claims. I don't do that, because there's nothing interesting, nor proofs no papers. There's usually rhetoric only and FUD.

[kraftman]

It's like talking with an idiot. I'm asking for papers which will backup what someone's claiming, but idiot gives links to sun people (who lie and FUD) blogs. I don't care if there are SOME papers, I want papers backing up your claims. The one which was worthy is obsolete. Now, next paper, please.

[energyman]

Sun did well techincally. From a business view point, Sun did not do well. Windows is better business thatn Linux (MS earns more money) - does this mean that Windows is better technically, than Linux? Of course not. Wrong.

let me rephrase it. Solaris was such a technological marvel that its marketshare was obliterated by linux on the server/enterprise/hpc front while Windows took away its client business.

Sounds like from a technological point of view Solaris had nothing to keep such customers. Maybe paying a shitload of money for a system that is dog slow didn't seem like such a great idea.

Top500. Jesus not that one again. I have posted ~10 times about top500 and explained how top500 does not count. Here is another explaination. At place 5(?) is Blue Gene. Blue Gene uses something like 700MHz PowerPC cpus. Now, according to your logic, 700MHz PowerPC is among the fastest CPUs in the world! Correct, or wrong?

your example and the conclusion is completely idiotic.

Supercomputers care about three things:
scalabiliy
performance
robustness

in no special order.

Solaris was once big in that business. A long, long time ago. Today? Noneexistant. Must be Solaris performance, scalability and robustness that made those people flock to it.

Read my earlier posts about top500. It says nothing. Super computers do only one thing well. Specialized hardware is easy to do. Specialized kernel is easy to do. Super computers are basically a network with lots of PCs. Linux scales well horizontally, but not veritaclly.

for someone claiming to be such an expert you know shocking nothing about supercomputers. Most supercomputers are build from parts you can buy. Commodity hardware. The time supercomputers were special built with special CPUs and special interconnects are GONE.
For almost 20years.

Sleeping well on all that SunOS tapes?

But according to you, Solaris must suck because it only runs on SPARC really well. A very... specialised CPU.

[energyman]

(again) http://bulk.fefe.de/lk2006/talk.pdf

ZFS did not look great back then. And Solaris hung. Yeah, that really makes me confident about Solaris' capabilities.

[LightningCrash]

What do big businesses use for their back ends?

Paypal has thousands of Linux boxes connecting to just a few large ______ boxes running the backend.

I'll let you fill in the blanks...

[energyman]

What do big businesses use for their back ends?

Paypal has thousands of Linux boxes connecting to just a few large ______ boxes running the backend.

I'll let you fill in the blanks...

IBM mainframes.

Everything else is not a 'big box'.

[LightningCrash]

IBM mainframes.

Everything else is not a 'big box'.

So you're saying an SGI supercomputer with 1024 CPUs is not a big box?