Torvalds Has Harsh Words For FreeBSD Devs

An anonymous reader writes "In a relatively technical discussion about the merits of Copy On Write (COW) versus a very new Linux kernel system call named vmsplice(), Linux creator Linus Torvalds had some harsh words for Mach and FreeBSD developers that utilize COW: 'I claim that Mach people (and apparently FreeBSD) are incompetent idiots. Playing games with VM is bad. memory copies are _also_ bad, but quite frankly, memory copies often have _less_ downside than VM games, and bigger caches will only continue to drive that point home.' The discussion goes on to explain how the new vmsplice() avoids this extra overhead."

Comments

Wrong Side of Bed?

[AKAImBatman]

Ok, let me see if I've got this straight:

• Copy on Write saves you real memory, cache memory, and CPU time by pretending that each forked process has a true copy of a memory segment when it in fact is looking at the original. That is, right up until a fork tries to write to that memory location, in which case an exception is handled by making an actual copy to a new location and allowing the write.
• Linus believes that the exception will occur enough in real world usage that it will be slower than just doing the copy in the first place.
• Linus wants to push the manual use of zero-copy memory sharing through the vmsplice() routine. He believes that the programmer will always know better than the system when to share memory.
• Linus doesn't like "VM Games" despite the fact that Virtual Memory, Memory Mapped Files, Disk I/O, Write Caching, etc, etc, etc, are all already "Memory Games" and "VM Games"

Do I have that right?

If so, I'm not really seeing his issue. Or at least not as hard-line as he sees it. The issue of memory copy performance is a tricky one, especially since CPU cycles are not the be-all to end-all of performance. Does the exception generated really cost that much more than he believes, or is it often eclipsed by the cost of the extra memory read/writes and CPU waits that are normally generated by a copy? Is it really feasible to expect program developers to do manual memory management in a day in age when programs easily weigh in at hundreds of megs?

I'm just not sure that Torvalds is really looking at all sides of this. He may be right, but I'd like to hear more discussion between the *BSD guys and Torvalds before we put this matter to rest. And preferrably without the insults this time. :-/

Links:

Copy on Write as explained by Wikipedia
FreeBSD page on Zero Copy Patches
Duke Uni Research

Re:Wrong Side of Bed?

[qwijibo]

I don't consider myself an expert in kernel programming, but I definitely think someone is off base if they're expecting programmers as a whole to do the right thing. Many programs seem to work by coincidence rather than design. People didn't do their memory management right in the days when it was necessary. Now that a lot of people are moving towards languages that handle the memory management for them, I expect even fewer to worry about it. That does mean that the programmers of the programming languages are the ones who are responsible, but I'd personally rather have the kernel take a more active role in memory management.

Re:Wrong Side of Bed?

[mrsbrisby]

Copy on Write saves you real memory, cache memory, and CPU time by pretending that each forked process has a true copy of a memory segment when it in fact is looking at the original. That is, right up until a fork tries to write to that memory location, in which case an exception is handled by making an actual copy to a new location and allowing the write.

No. Updating the page tables twice and having a fault in there is very expensive.

Linus believes that the exception will occur enough in real world usage that it will be slower than just doing the copy in the first place.

And he's right too. But he's not recommending the copy "in the first place" - he's recommending explicit notification that the pages aren't used anymore instead of an implicit notification by-way of a page fault.

Linus wants to push the manual use of zero-copy memory sharing through the vmsplice() routine. He believes that the programmer will always know better than the system when to share memory.

That's correct.

Does the exception generated really cost that much more

Yes. There isn't a grey area on it either- it's basic math: cost of page copy + exception + 2 * (page table update) is greater than cost of page copy + page table update.

The real issue is that the userland knows what it's doing. Eventually it'll want to reuse a buffer. Now does the userland start reusing pages when malloc() fails- thus incuring the exceptions when memory is tight? Or does it reuse them when the kernel says they're reusable?

The latter makes more sense if you're actually concerned about performance. The former may be easier to code, but I doubt many people will actually do that because it's hard to test.

In practice what people do is use a static buffer- that's even EASIER to code, but it means page faults happen ALL the time.

Is it really feasible to expect program developers to do manual memory management in a day in age when programs easily weigh in at hundreds of megs?

They already have to do it. Whether it's the BSD implementation or the new Linux implementation they already have to do it if they want reasonable performance in the real world.

To really take advantage of the BSD implementation, your program needs to monitor malloc() usage, and start attempting to reuse pages when it fails- oldest to newest. This is complicated and hard to test.

To really take advantage of the Linux implementation, your program waits until it gets notification (via select() or poll()) on the vmsplice() recvmsg() operation. Once that occurs, the notification says exactly which pages can be used.

The result? Userland on Linux is easier to write, and easier to test. It'll also be faster.

Re:Wrong Side of Bed?

[RailGunner]

Yes, you've got it right on what Linus is saying.

The issue of memory copy performance is a tricky one, especially since CPU cycles are not the be-all to end-all of performance. Does the exception generated really cost that much more than he believes, or is it often eclipsed by the cost of the extra memory read/writes and CPU waits that are normally generated by a copy? Is it really feasible to expect program developers to do manual memory management in a day in age when programs easily weigh in at hundreds of megs?

What programs weigh in at hundreds of megs? Don't count data files or map files for games. The entire bin directory of a PostgreSQL install is only 20 megs, and that's a lot of stuff there.

And as far as doing memory management... YES. I have yet to see a compiler do a better job at managing memory than what I can do when I write my code - and the reason is quite simple: I'm the domain expert, not the compiler. Compilers generally do a good job, but it's those specific cases that bite you over and over again.

Linus is also right about child threads writing to memory. If that never happened, we wouldn't have a concept of a lock or a semaphore. The bottom line is that is happens a lot.

He may be right, but I'd like to hear more discussion between the *BSD guys and Torvalds before we put this matter to rest. And preferrably without the insults this time. :-/

I agree, the ad hominem was completely unecessary.

Re:Wrong Side of Bed?

[mrsbrisby]

One thing that concerns me about making all of these copies is that it seems like a quick and easy way to blow out your L2 cache. That could in the long run have a worse performance penalty than having to play the VM tricks with CoW.

No it won't. The only way to avoid the copies is to avoid the pagefaults. Since userland doesn't get explicit notification in FreeBSD of when the pages are safe to use, the process should wait as long as possible (e.g. until malloc() starts failing)

The idea Linus pushes here is explicit notification- via select() or poll() and returnable via recvmsg(). That way the userland knows exactly which pages can be reused.

The result is that it's faster and easier to develop userland programs to take advantage of it. It's also easier to degrade gracefully into read()/write() until the FreeBSD people see the light and add support for this too.

It's really a clever idea.

Re:Wrong Side of Bed?

[visgoth]

Because everyone enjoys a good old fashioned jihad once in a while?

Re:Wrong Side of Bed?

[Nato_Uno]

He doesn't care what the FreeBSD developers are doing... ... until someone advocates copying their ideas into the Linux kernel. Then he cares very much.

He's not saying "The FreeBSD people should rewrite that part of their OS," he's saying "don't put that crap into the Linux kernel."

Re:Wrong Side of Bed?

[tearmeapart]

>> Linus wants to push the manual use of zero-copy memory sharing
>> through the vmsplice() routine. He believes that the programmer
>> will always know better than the system when to share memory.
>
> That's correct.

No, that is not always correct.

I am a C developer for a large multinational corporation that likes to make money. When I need to fork(), I do not have the time to think of all the memory management invovled with fork(). I just want it to be done reliably, and I want it to be done fast.

If it turns out that my code runs 10% faster on FreeBSD than on Linux, than that means that the code is probably going to go on a FreeBSD system. And if FreeBSD is not an option, than I am not going to do the optimization (because CPUs cost less than my wages).
Also: optimization never happens anyways (or at least, not properly).

So from my perspective:
I want the kernel to run my code as fast as possible by default.

Re:Wrong Side of Bed?

[AKAImBatman]

If you use a static buffer, you ALWAYS cause a fault.

* Lightbulb goes on

Oohhhh, I see! So something like this is the problem:

char buffer[1024];
int read = 0;
int length;
 
while(read < totalSize)
{
    length = fread(buffer, 1, 1024, &file);
    read += length;
 
//Do some stuff, but don't free the buffer!
}

What you're saying is that every time through the loop, there's going to be a page fault as the CoW pages are wiped away by the new copy into the same logical buffer. CoW is dependent on allocating new pages every time so that you don't ever write to the old CoW pages. Correct?

Of course, this is where I'd really like to hear from the *BSD developers. Surely they must be aware of this issue? Do they expect programmers to throw away their buffers, or do they have a plan?

Re:Wrong Side of Bed?

[LordNimon]

I don't consider myself an expert in kernel programming, but I definitely think someone is off base if they're expecting programmers as a whole to do the right thing.

Well, I am an expert in kernel programming, and I can tell you that Linus has little tolerance for anyone who doesn't program the way he does. That's one reason, for example, that he doesn't support debuggers. Every other OS has a kernel debugger built-in (and therefore, generally stable and full-featured), but not Linux. Even the OS/2 kernel debugger that was created 10 years ago is better than anything Linux has.

Re:Wrong Side of Bed?

[mrsbrisby]

What you're saying is that every time through the loop, there's going to be a page fault as the CoW pages are wiped away by the new copy into the same logical buffer. CoW is dependent on allocating new pages every time so that you don't ever write to the old CoW pages. Correct?

Exactly correct. Those frequent CoW operations are slow- the page faults are expensive. If you had instead written:

        char *buffer;
        int read = 0;
        int length;

        while(read < totalSize)
        {
                buffer = malloc(1024);
                length = fread(buffer, 1, 1024, &file);
                read += length; //Do some stuff, but don't free the buffer!
        }


Then it would operate quickly on FreeBSD. The problem then becomes exactly when do you free all those malloc()s?

On Linux, you can get a signal from the kernel- via a recvmsg() call that will tell you exactly which pages are now available to be freed- or better still, reused.

It'll be easy to check and test correctness AND the programmer has to be aware it's going on in order to use it at all.

Under FreeBSD the programmer can use the syscall, but never get the performance unless they know exactly what's going on.

Of course, this is where I'd really like to hear from the *BSD developers. Surely they must be aware of this issue?

I don't know. The article wasn't about that- I doubt Linus pays attention to what the BSD people know- in fact, I don't even think he knows for certain if FreeBSD even works this way. :)

The point is that using CoW is stupid for this. It makes things complicated in the hard case, and in the easy case, it makes things slower.

Re:Wrong Side of Bed?

[mrsbrisby]

But that's not true in general. 99% of all fork() calls are followed by exec() and the entire space gets dumped. That's why COW is a huge win in the average case. The case of an application using fork() followed by actually doing something useful is exceptionally rare outside of the server space. In fact, Apache is about the only program I can think of that ever does this.

This isn't about fork() it's about zero copy buffers, not code and data pages in general.

Consider a block like this:

char buffer[4096];
for(i = 0; i < len;) { r = read(fd, buffer, 4096); zero_write(fd2, buffer, r); i += r; }

Now, on the whole, if zero_write() works like write() then an awful lot of copying is going on. But if zero_write() uses the buffer for kernel space as well, it's much faster (1 less copy).

Now the trick is returning to userspace before the buffer is completely used. In FreeBSD a page fault would occur immediately during read().

Both FreeBSD and Linux agree that you shouldn't do this. Instead something like this:

char *buffer;
for(i = 0; i < len;) { buffer = malloc(4096); r = read(fd, buffer, 4096); zero_write(fd2, buffer, r); i += r; }

The trick at this point, is that elsewhere in your code, Linux can tell you when those malloc() buffers can be reused, whereas FreeBSD doesn't. It relies on the fact that you'll either make a blocking call on fd2 before you free buffer _OR_ you'll accept a page fault.

But if you can be told when it will occur, you don't need to do either of these things, and as a result, you NEVER have to wait. This means your program will be simpler and go faster.

Re:Wrong Side of Bed?

[Jherek Carnelian]

When I need to fork(), I do not have the time to think of all the memory management invovled with fork().

This has NOTHING to do with fork(). You are used to CoW (copy-on-write for anyone else reading along) only applying to fork(), but that is not the issue under discussion at all. You, and probably 95% of the responders here, need to go RTFA.

The issue is implementing zero-copy IO. FreeBSD's way of doing it do a setsockopt() that causes any write() on that socket to mark the buffer CoW so that it can use it exclusively for handing down to the device driver. The "magic" is that if the programmer tries to use that buffer while the device driver owns it he will get a copy. BUT, the programmer has no way of knowing when that buffer is available again.

Linus's point is that marking a page CoW is very expensive - especially in an SMP environment, almost as expensive as just copying that page to begin with would be. He also argues that taking a page-fault to invoke the CoW to a new page, or simply to turn off the CoW attribute, is orders of magnitude more expensive than just copying it in the first place.

So that means the CoW for sockets is only really useful if you rarely or never reuse your buffers again. And the only place that happens is in synthetic benchmarks.

If Linus had said "Microsoft is a bunch of idiots for implementing a feature that only looks good on benchmarks" everybody would be nodding their heads in agreement. I think the reason people are not doing the same here is because they just don't understand the details.

Re:Wrong Side of Bed?

[outZider]

Here's my -1, Troll.

Funny that we just had an article about how many Linux users and enthusiasts exclude other people by being complete dicks, and here you are, acting like a dick. Of course, I don't know you from Joe Blow, so maybe I just misunderstood your obviously angsty response.

"That's obvious."
"Correction: when _you_ start using up a lot of memory Linux totally sucks. When I start using up a lot of memory, Linux acts exactly as I expect, and better than FreeBSD."

Linux acts exactly how I'd expect, too. It completely sucks when it comes to memory and process management. Linux may have a better threading kernel, but that's the only thing that seems to save it in the real world. After only six years of administering servers professionally and for my own use, it has come down to Linux on the desktop, and FreeBSD for Real Work(tm). Many large companies that depend on their data agree with me, and those who use Linux or Windows just throw more machines at the problem.

At least Linux is free compared to Windows, right?

Re:Wrong Side of Bed?

[Just Some Guy]

No BSDer has been willing to reproduce the tests, as it will only confirm what the marketplace has already decided ... Linux is the superior OS.

...but still inferior to Windows, right? I mean, we're only looking at number of installations, after all. Furthermore, McDonald's clearly has the best hamburger and Velveeta beats Hoffman's Super Sharp.

I like Linux. I'm typing this on a Gentoo box. However, I'd never pretend that it's better in every single aspect than any other OS in existence. The BSD guys have a few tricks up their sleeve, and even Redmond manages to get things right on rare occasion.

Tantrums

[ClamIAm]

Methinks we need to start tagging "tantrum" to this type of thing.

Given the respective quality of the Linux and *BSD

[Anonymous Coward]

kernels, me thinks it's just sour grapes because Linus can't compete in that area.

Wrong side of compiler

[StarKruzr]

I think Linus has gotten to the point where he just really enjoys trolling. Like, this was OBVIOUSLY uncalled-for, and he's usually such a laid-back guy. Maybe's he's read too much Slashdot. I don't know.

Re:Wrong side of compiler

[nuzak]

Actually he's been into boorish behavior from day 1 when it comes to microkernels. Namecalling between him and Tanenbaum (admittedly Tanenbaum is a bit haughty and provoking), and his slanderous accusations against microkernel researchers in general (a quote I can't find at the moment, but he basically accuses them all, as one big class, of academic fraud to procure grant money).

The only microkernel Linus knows jack about is Mach, an ancient piece of crap, which indeed is Linus indeed calls it. It's unfortunate real-world systems were saddled with it, and it's got real performance issues, but Linus carries on about it like Mach ran over his dog or something.

He conveniently ignores or chooses to remain ignorant of the fact that L4Linux is typically faster than Linux itself. To say nothing of the real-world success of QNX. And even L4Linux is pretty old by today's standards.

This is all pretty typical behavior of Linus: bluster now, bone up and learn, and implement it later. He did so with SMP (saying famously that the way to do it was one Big F**ing Lock, then learning that no this wasn't such a great idea after all). Then he went on a tirade about sun's /dev/poll before learning that yes they actually didn't cheat and they did it smarter, and Linux followed.

Ultimately, Linus and Linux come around. Sometimes he just has to vent.

Re:Wrong side of compiler

[arivanov]

More likely he had some really bad acid the previous night.

After all he did more than 6 revisions of the Linux VM using CopyOnWrite before this latest fad.

Possibly more.

Off the top of my head that is at least 1 in the 1.2 tree, 1 in the 2.0 tree, 1 in the 2.2 tree, 2 in the 2.4 tree and more than 2 in the 2.6 tree, all of which being CopyOnWrite and at least some of which has been hailed as the next best thing after hot bread.

As far as the technical point he is possibly correct for x86 where COW goes through the fault mechanism and causes some TLB and cache abuse which is really bad on modern CPUs. I am not sure as far as other architectures are concerned, because IIRC (I may be wrong) the memory mapper hardware on the old Sparc was designed for COW in first place.

Anyway, before calling somebody else an idiot for something you have happily done for 10+ years till yesterday it may be nice if you look at yourself in the mirror. Because I never remember any branch of FreeBSD reaching the point where you can do a find /usr -exec cat {} > /dev/null \; to hang the system. That is 2.6.16 at your service (from rc4 onward) on at least two x86 subarchitectures where I had the time to test it. That is besides the unkillable processes in [S] state on an nfs flock in 2.6.14 (yep, that is a gem which no other unix has managed so far), besides the OOM idiocies in 2.6.10, besides deliberately making it absolutely impossible to backtrack any more interesting patch to a previous kernel without employing a team of kernel developers because the VM and locking is not compatible across any kernel version since 2.6.9 and even when it is something else is changed like the tty layer, besides.... Aarghh.....

Re:Wrong side of compiler

[bfields]

I think Linus has gotten to the point where he just really enjoys trolling.

Could be:

I got slashdotted! Yay!

On Thu, 20 Apr 2006, Linus Torvalds wrote:
>
> I claim that Mach people (and apparently FreeBSD) are incompetent idiots.

I also claim that Slashdot people usually are smelly and eat their
boogers, and have an IQ slightly lower than my daughters pet hamster
(that's "hamster" without a "p", btw, for any slashdot posters out
there. Try to follow me, ok?).

Furthermore, I claim that anybody that hasn't noticed by now that I'm an
opinionated bastard, and that "impolite" is my middle name, is lacking a
few clues.

Finally, it's clear that I'm not only the smartest person around, I'm also
incredibly good-looking, and that my infallible charm is also second only
to my becoming modesty.

So there. Just to clarify.

Linus "bow down before me, you scum" Torvalds

Re:Wrong side of compiler

[kv9]

Finally, it's clear that I'm not only the smartest person around, I'm also incredibly good-looking, and that my infallible charm is also second only to my becoming modesty.

i know what's the last book that linus read. do you?

Re:Wrong side of compiler

[nuzak]

> what the? L4Linux has to run on top another REAL kernel, usually Linux.

You're quite mistaken. L4Linux runs Linux in usermode on top of the L4 kernel.

http://os.inf.tu-dresden.de/L4/LinuxOnL4/

Linus, my advice to you is simple...

[Anonymous Coward]

.. this will help you keep yourself calm.

Sweet

[bogie]

It's been a while since we had a huge linux vs BSD flame feast.

I'll start.

BSD user: Linux is a confusing mess of programs and is less stable than BSD.

Linux user: Your still here? I thought you were dead by now?

Linus sometimes calls people idiots

[cpu_fusion]

And in other news...
Grass is green;
Oil is overpriced;
Absolute power corrupts absolutely.

Re:Linus is turning into a dictator

[Lumpy]

No he is simply getting less tolerant of "sloppy" programming. He is one of the very very few that believes in doing it the way that gives you the best speed. Something that takes 4+ operations compared to a way of doing it with only 2 operations and you get less problems = performance gains that add up. Just because your typical machine has 4 dual core 8Ghz processors and 22 terabytes of ram does not mean you can slack off and write the whole thing without paying attention to performance.

the BSD guys have their reasoning and if you read more info about this it is not a shot in the dark that Linus is taking but he is frustrated that after many discussions nobody cares as much as he does on the performance issues.

Go back and read what Linus did back in the early days, it's no different today than what it was in 1990, he will call a duck a duck.

The Universe In Which Spock Has A Beard?

[tlambert]

``And in what universe is anyone who can intelligently speak about (much less code around) memory and VM management [be called] an "incompetent idiot"?''

The Universe In Which Spock Has A Beard?

-- Terry

Re:Linus is turning into a dictator

[rtaylor]

No he is simply getting less tolerant of "sloppy" programming.
You'll forgive me for taking that with a grain of salt so long as memory over-commit remains the default mode of operation within Linux.

RTFA, please. Or at least my summary here.

[ColonelPanic]

The complaint is not about general copy-on-write, it's about BSD's ZERO_COPY_SOCKET feature vs. vmsplice().

Basic explanation: Suppose that a program is doing a lot of output to a file or socket. The program can generate data faster
than the kernel can consume it, say. So what should the kernel do with the buffer it receives from the user on each write()?
There are three options.

1) Copy its content immediately elsewhere, so that on return to User Mode, the buffer remains writable and writes are safe.

2) Change the access rights of the page containing the buffer, so that no copy need be made unless User Mode attempts
      to modify its content before the kernel has completed the write(). If the user attempts to write, it either gets
      permission to do so (because the kernel is done) or it gets a writable copy.

3) Let User Mode promise to not modify the buffer's content until told that it's safe to do so, leaving it writable in
      the meantime.

The default behavior is (1); BSD's zero copy socket feature is (2), and the point of Torvalds' complaint; vmsplice() is (3).

What harsh words?

[Inoshiro]

Andy went out and said that he thought the Linux approach was wrong, and archaic, and that people should go and wait for GNU.

Linus said that he felt this was wrong, and that being a prof is no excuse for Minix being the mess it was (and Minix was a mess in the late 1980s/early 1990s). He also apologized if he came off as too harsh for his writing about how people should be able to throw away an old design in favour of a new one anyway, etc.

It was very polite compared to some of the non-Andy/Linux replies.

An explanation

[sjames]

There seem to be a LOT of misconceptions about the discussion of vmslice() vs COW vs copy. This has nothing to do with conserving memory and everything to do with high performance I/O. If your app just needs to send a couple small files from A to B, you probably don't care about this at all.

A little background is needed on the terminology and mechanisms of I/O for any of this to make sense. For an example, let's say your app is a very busy web server sending dynamic (but trivial to compute) pages out.

The oldest and simplest method is copy. The app calls write(int sock, char *buffer, int length) on a socket. The kernel coppies the contents of buffer from userspace memory into a kernel space buffer and at least queues the data to the TCP stack before returning.

COW is an attempt to avoid the cost of copying the outgoing data.. In that case, the reference count on the physical pages that make up buffer is bumped up (since now kernel and application are both interested in them), and marks the pages as COW. That is, the virtual memory addresses are set as read only and a flag bit is set (more or less). The latter is done so the kernel needn't worry about them again. By the time the write call returns, the app is able to immediatly write to that memory (sorta) without worry.

When that write happens, the app takes a page fault (writing to a read-only page). The kernel sees that the pages are COW, copies the data to a new physical page, and maps the page in read/write. Then it returns from the fault. OTOH, if the kernel finished with the page first (the data goes on to the wire), it re-marks the page(s) so the app can access them without a copy.

The hope is that often enough, the app WON'T try to write to the pages while they're busy and so the cost of that copy is saved. If that hope comes through often enough it MIGHT be vaguely uesful. I say MIGHT since there is a significant cost just for marking the pages (the CPU's TLB must be flushed for the change to take effect). If the faults happen, it's a BIG loss since handling a fault takes thousands of CPU cycles.

So, for it to have any chance to help, the application programmer must already know enough to TRY to avoid writing to the same buffer again until it gets to the wire. Unfortunatly, it can never be sure so most apps don't bother.

The vmsplice() proposal is fairly simple. In this case, the app explicitly requests special treatment of the write. The pages are NOT marked as read only at all. Instead, the app is on it's honor to leave them alone until the kernel notifies it that they are again available. This saves the copy and the costs of TLB flush AND the (potential) cost of page faults. If the app breaks it's promise, it is the only one to suffer as the data it sent is corrupted (no kernel housekeeping is ever stored in such pages so there are no security implications). Any damage the app might do by sending screwy data could also be done using the old copy method.

What it all comes down to is that playing tricks with page mapping LOOKS nice at first glance since it SEEMS reasonable that not copying bytes around will save CPU cycles and memory bandwidth. The re-mapping (or just permission changes) on pages SEEMS lightweight. Unfortunatly, in fact, re-mapping or changing permission forces cache invalidations and page faults are just plain expensive. With the direction CPU design is going, these things will likely get more expensive rather than less (as they have for most of the history of microprocessor design).

It's really not that complex for an application to use. At least in comparison to the complexities and level of knowledge required to write an app that performs well enough to need this in the first place.