[waffffffle]

I am in an operatings systems class right now and the professor today spoke about the different kernel approaches. He showed the pros and cons of both the microkernel and the monolithic kernel idea. he stressed that a microkernel makes the os extremely slow because of all the jumps between user and kernel space and the memory space for all the system services. NT and linux both use monolithic kernels, which also has its downsides, but it have the advantage of speed.

So if Apple had gone with a different unix as a basis for the current Mac OS, with a monolithic kernel, would OS X be faster?

[El Pre$idente]

I just want to know why all browsers suck.

[voodoo]

What are the benefits of a microkernel? Why would one choose that as a core for an OS?

[snerdini]

Originally posted by waffffffle:
...linux both use monolithic kernels, which also has its downsides, but it have the advantage of speed.

That explains why Linus Torvalds expressed his dissatisfaction with OS X and the Mach kernel...

[Mactoid]

Originally posted by voodoo:
What are the benefits of a microkernel? Why would one choose that as a core for an OS?

It's modular, and more flexable. Allows for a much smaller, cleaner code base. Doesn't require entire kernel recompiles for many changes. It is a just plain better design.

Of course, it can be argued that despite it's benefits, the monolithic kernel is better in practice because of it's speed benefits. This is yet another topic of major Holy Wars.

[someone_else]

Originally posted by voodoo:
What are the benefits of a microkernel? Why would one choose that as a core for an OS?

First of all, to respond to the OP, the OSX kernel is not a micro-kernel as your professor explains it. In OSX, the BSD "personality" (in Mach terms) is not a user-space daemon as it would be with a true Mach micro-kernel (such as GNU Hurd). The BSD kernel and the Mach kernel are compiled into the same binary (/mach_kernel) and share the same address space. IOW, there is no true split between the BSD and Mach kernels and you can think of the OSX kernel as monolithlic in the same sense that Linux is. So in answer to your question; no, Mach is not the reason for OSX slowness. The "slowness" as you call it is much more a hardware problem (very pore memory throughput, video cards that aren't tuned for Quartz, and yes low Mhz CPUs) than a software problem.

Now for the benfits of Mach in OSX.

1. Abstraction - Mach provides a hardware abstraction layer to the BSD kernel portion. It uses a very well defined API (unlike BSD) and makes hardware changes much easier.

2. Threads - Mach has threads at it's core. BSD does not have native threads (at least until FreeBSD 5.0 goes final).

3. Realtime support, for Audio, Video, etc. Again BSD does not have RT support.

4. Mach ports. These are native kernel communication channels and make things like Apple Events on OSX possible. W/O Mach and Mach ports, OSX would be a very different (and much less "friendly") OS.

This is just the highlights. Mach provides a lot of functionality that BSD does not.

HTH.

[MacGorilla]

The mach microkernel has nothing to do with the speed of Mac OS X. mach handles very low level I/O stuff and interacts with the hardware on a byte level. It communicates with the darwin kernel through a native messaging system.

[Green Leaf]

screw you man
my computer is fast

how dare you say that OS X is slow
what do you know about computer
what ever it is
i know more

HEY EVERYONE
this guy is ripping on apple
therefore we have to band together
and deny these allegations


who do you think you are
turtle breath

[MacGorilla]

Originally posted by Green Leaf:
screw you man
my computer is fast

how dare you say that OS X is slow
what do you know about computer
what ever it is
i know more

HEY EVERYONE
this guy is ripping on apple
therefore we have to band together
and deny these allegations


who do you think you are
turtle breath

Go back to whatever hole you just crawled out of Kelly Hogan.

[Mactoid]

Originally posted by snerdini:
That explains why Linus Torvalds expressed his dissatisfaction with OS X and the Mach kernel...

True. Linus is totally opposed to Microkernel architecture. In fact, this was the topic of a legendary debate between Linus and Andrew Tanenbaum, a well respected researcher in the field of kernel design.

[someone_else]

Originally posted by MacGorilla:
The mach microkernel has nothing to do with the speed of Mac OS X. mach handles very low level I/O stuff and interacts with the hardware on a byte level. It communicates with the darwin kernel through a native messaging system.

No, Mach and BSD are both the Darwin kernel (see my previous post). The two kernel entities communicate through function calls, not message passing. There is no need for message passing, since they share the same address space.

[intastella]

Take this with a grain of salt. I'm not a developer, but I have read up on alot of these issues.

But the short answer is no.

The slowness you speak of is still just a side effect of Quartz. It has lots of overhead and uses the processor for most of the operations, where the graphics card did most of this in the past. Quartz Extreme helps this but still doesn't solve it completely. What you get for this trade off is an amazing graphics system, basicly Adobe AfterEffects compositing on the fly.

As far as the kernel is concerned, Mach has evolved alot while at Apple, and the messaging - microkernel that you probably heard about is more of the pure microkernel that Mach was back in the day. The Mach that is in OS X is a bit of a hybrid, as it incorporates some elements that were seperate from the kernel in the past. It does this for more speed and lower latency.

This is all user speak for complicated issues but I think it should allay some concerns.

[fmalloy]

I suspect that in reality, much of OS X's slowness is the complexity of the Quartz compositing layer.

[MindFad]

Originally posted by MacGorilla:


Go back to whatever hole you just crawled out of Kelly Hogan.

LoL

Hey, I just bought a new HD and installed Jaguar to make my full transition soon, and I must say, Jaguar is way faster than 10.1.5 I had been using for months. Quartz Extreme is a godsend. Scrolling is vastly improved (and not from QE) -- I actually enjoy browsing the 'Net (with Chimera) now since scrolling is so much better. It's better, but it can still be even better. When scrolling in Photoshop is like it is in OS 9, I will be a very happy man. Until then, I need more RAM and patience with Apple. And bring me a Power4, Apple.

[MacGorilla]

Originally posted by someone_else:


No, Mach and BSD are both the Darwin kernel (see my previous post). The two kernel entities communicate through function calls, not message passing. There is no need for message passing, since they share the same address space.

Ahh...thanks for the clarification!

[macmike42]

Originally posted by fmalloy:
I suspect that in reality, much of OS X's slowness is the complexity of the Quartz compositing layer.

<RANT>

Other than resizing windows (try forcing any other OS to completely redraw the contents of the window before repositioning the window frame to match the mouse position and you'd get the same result), what slowness are we talking about?

I am running a beige G3/300 with a lowly Rage 128, and I would say that on average, 10.2.1 is perhaps 5-10% slower than OS 9.2.2. Many operations are considerably faster. This is not bad considering that this is the lowest end machine that supports OS X (it's really a rev. a 233 overclocked to 300).

</RANT>

<FACTS>

As for the Darwin kernel, here is what I understand:

Darwin is a hybrid kernel based on Mach which use monolithic design ideas.

Kernel extensions load directly into kernel memory space. In a "true" microkernel, nothing can load into the kernel's memory space except the kernel. All drivers exist as servers in user space, and all communication with the drivers must be done via Mach messages. This is much slower than function calls and pointer passing within kernel address space, which is the way the Darwin kernel operates.

As for the BSD side of things, you can rest assured that no microkernel would ever try to emulate a UNIX-like operating system, as this would sacrifice stability. Things such as virtual memory, file I/O, system calls, and IPC would be handled by user space servers. Darwin, on the other hand, handles all of these thing internally (in kernel space), and even handles both POSIX IPC and Mac OS AppleEvents as Mach messages natively.

Hopefully at this point the layperson can begin to understand that:

1) Microkernels truly are slow, but they were designed for stability, not performance.

2) Darwin is a Mach-based kernel that combines ideas from both the BSD and Mach kernels, arguably* getting the best of both worlds.

</FACTS>

* I say "arguably" because Darwin is still a young kernel, and I can crash it with two lines of code (in a kernel extension). I have seen vast performance improvements since the public beta and only one kernel panic. It isn't quite as fast as Linux, but it is getting close.

[funkboy]

Originally posted by Green Leaf:
screw you man
my computer is fast

how dare you say that OS X is slow
what do you know about computer
what ever it is
i know more

HEY EVERYONE
this guy is ripping on apple
therefore we have to band together
and deny these allegations


who do you think you are
turtle breath

Wow, it's a poem... like e.b. cummings!

I don't think the kernel choice will slow down an OS *that* much. If you're running at 1 MHz or 2MHz, yes, you will notice a difference between kernels. But I somewhat doubt you will on a 500MHz machine... many other bottlenecks come first, though I suppose one could be all the writing to and from the hard drive.

[snerdini]

Originally posted by funkboy:

Wow, it's a poem... like e.b. cummings!

Isn't it e.e. cummings?

[funkboy]

Originally posted by snerdini:


Isn't it e.e. cummings?

I just found out the Miss Teen USA from my state's favorite author was E.B. White, I must've been a bit confused... you're right

(I love E.B. White, too... his book of essays is sitting right in front of me...)

[Millennium]

Originally posted by snerdini:


Isn't it e.e. cummings?

In the year 2057, geneticists create a mutant hybrid clone of e.e. cummings and E.B. White, to write poems on spider webs. This is his story...

[Millennium]

Originally posted by someone_else:
No, Mach and BSD are both the Darwin kernel (see my previous post). The two kernel entities communicate through function calls, not message passing. There is no need for message passing, since they share the same address space.

It's worth noting that neither NeXT nor Apple "added" BSD to the Mach kernel. It's actually in there "normally", by design. As such, it's not really a true microkernel.

There are some Mach-based kernels which take out the BSD layer. But these are derivatives of Mach, not the other way around.

[mitchell_pgh]

I hear that the big issue regarding speed is the left over debug code in the kernel... From what Slashdot said, it should be removed in 10.2.2

[godzookie2k]

Originally posted by mitchell_pgh:
I hear that the big issue regarding speed is the left over debug code in the kernel... From what Slashdot said, it should be removed in 10.2.2

please tell me you are joking.

[Mactoid]

Originally posted by mitchell_pgh:
I hear that the big issue regarding speed is the left over debug code in the kernel...

[Phanguye]

ha some rumors just wont die... even when proven to be absolutely false

[Mongrel]

oh man... this reeks of Piracy-bait...

[Kristoff]

When Linus figures out threading, then he can give his opinion.

As for the rest of it...MACH in its essence is a microkernel....DARWIN is not.

All this is very well explained on pages at CMU and APPLE....try google folks!

[someone_else]

Originally posted by Millennium:

It's worth noting that neither NeXT nor Apple "added" BSD to the Mach kernel. It's actually in there "normally", by design. As such, it's not really a true microkernel.

There are some Mach-based kernels which take out the BSD layer. But these are derivatives of Mach, not the other way around.

That was true of Mach 1.x and 2.x, but not Mach 3.0. Mach 3.0 was/is a true micro-kernel that had the BSD layer implemented as a user land daemon (HURD does it this way, although I don't think they use BSD).

So your statement is true for NeXT, but not for Apple. NeXT never shipped with the 3.0 kernel, Apple has.

[Northform]

A microkernel isn't slower nor is a monolithic kernel that unmodular. Loosely speaking, using a different kernel won't make an OS better. Sure, if one is complete crap...

Almost everyone believes that quasi-microkernels are much better than monolithic kernels. Why would the GNU organization be working on a microkernel if they thought that the philosophy behind the Linux kernel was better (yeah, bad reasoning, I know). Mac OS X doesn't use a true microkernel as such a system only incorporates the minium needed to run in it.

In the future (or today's high end) you can see why a microkernel design is better. Monolithic kernels (Linux) are unaware of threads and MP (hmm...what company seems to be relying on that).

<metaphor>Why do people use oop? Languages like C still work.</metaphor>

Back to topic: Monolithic kernels really have no advantage. Microkernels have advantages, but not really with today's consumer hardware.

BTW Windows XP is microkernel based NOT monolithic.

[Gee4orce]

Actually, choosing the architecture on the basis of speed is, in the long term, the wrong approach. Moores Law tells us that the hardware will always catch up to the greater demands of the software. Ultimately, the difference between microkernel and monolithic architecture speed will be imperceptible.

[asagoo]

Whether a kernel is good or bad can't be measured absolutely. Every design decision, such as modularity, has a trade-off, such as overhead. This is especially the case in OS design. You have to make these decisions based on what your needs are, and Apple has made the right decision because it makes it easy for them to switch hardware, use messaging, and all of the reasons mentioned before.

Another example: Solaris has a complex threading technique, which seems work well for them, but the designers of the new Linux thread implementation find that approach inappropriate.

As for the speed in Mach, I don't think that is something we have to worry about. 10.1 and 10.2 have proven that most of the speed improvements in Mac OS X are to be made at a much higher level, i.e. in Quartz, application launching, threading, etc.

By the way, another advantage of Mach, if I remember reading correctly, is that its memory paging mechanism is meant to be quite badass. That is assuming that Apple didn't replace the default pager with their own one.

Amar

[sniffer]

Originally posted by Northform:

BTW Windows XP is microkernel based NOT monolithic.

Thanx for clearing that up. It was what I thought too.. BTW: I also assume you includes all the NT versions under catagory; microkernel?

[dfiler]

Originally posted by Gee4orce:
Actually, choosing the architecture on the basis of speed is, in the long term, the wrong approach. Moores Law tells us that the hardware will always catch up to the greater demands of the software. Ultimately, the difference between microkernel and monolithic architecture speed will be imperceptible.

I think this applies to most software but not to kernel design. Other software is generally finite in resource requirements. A bad design might by twice as slow, but the task in question is finite. A kernel's performance is a different game entirely. A kernel has an effect on the speed of pretty much everything occuring in the computer. This means that as we find new things to do with the processing power of faster computers, these new things will also be effected by kernel performance. The additional processing overhead of microkernels is not a fixed figure which will become insignificant on faster machines. Rather, microkernel overhead will increase in a near linear manner with respect to processing power. Some of this can be alleviated with more sophisticated scheduling to cut down on message passing and mode switching. However, microkernels will still be slower no matter what. With that said, speed isn't the only concern is system design. Final note: OS X does not have a microkernel.

[sniffer]

Originally posted by dfiler:


(..)

Final note: OS X does not have a microkernel.

So OS X's kernel is an hybrid then? This is confusing. Thorvald Linus seemed to think about it as an microkernel in his book if I remember correctly.. But that was a time back OTOH..

[Nonsuch]

Originally posted by sniffer:


So OS X's kernel is an hybrid then? This is confusing. Thorvald Linus seemed to think about it as an microkernel in his book if I remember correctly.. But that was a time back OTOH..

Linus never directly mentions Mac OS X in his book. The whole "Torvalds bashes OS X" thing was blown way out of proportion and was little more than flamebait.

Here's what Linus said after the controversy:

Linus Torvalds said April 8 that reports of him trashing the new Mac OS X are simply misquotes, as he has "never commented on OS X." The reports stem from early looks at his upcoming autobiography, Just for Fun: The Story of an Accidental Revolutionary, which he coauthored with David Diamond.

The book does contain some criticism of a microkernel design called the Mach kernel, upon which the Mac OS X kernel is based. As a result, Torvalds has taken heat from Mac faithful. In an email interview with LinuxWorld.com Torvalds said: "Interestingly, the thing erupted over a fairly small passage in the book: it's not as if I spend a few hundred pages bashing Mach. In fact, I spend a lot more pages in the book complaining about how people always take me (and other 'tech luminaries,' for that matter) too seriously and get all hung up about details."

As for the Mac OS itself, Torvalds feels that its new kernel is a big step forward. "All of the interesting stuff in OS X is actually outside the kernel: what makes Macs special is not the OS (which historically has had tons of design problems, even more so than Mach ever has -- so OS X is bound to be a big leap forward), but the UI and the ease of use. Which has very little indeed to do with Mach," he said.

A lot of people do take Linus very seriously. I enjoy reading Slashdot forums where some young Linux kid posts about how microkernels suck and are slow, and then proceeds to get the sh!t flamed out of him for pages afterward.

[passmaster16]

This is a very intersting post as I am not familiar enough with *nix to really give any insight. I do agree with the notion that the slowness in X is attributed to the Quartz compositing system. You can see this in other *nix distributions as well when using certain window managers like KDE. I once loaded Linux on a 166 pentium and was trying to figure out why it seemed so slow. I asked our Unix admin at work why Linux was running so slow when it supposedly runs fast even on lower speed hardware. He responded that it's not Linux that's running slow, it's KDE that makes it appear to run slow; he went on to say to try running linux from the command line and you'll know what I mean. So I would have to agree with him that when running OS X on half decent hardware, the slowness is probably due to aqua. And I would say that this may be a good thing because the aqua engine still has room for improvement and will become faster in future releases.

[snerdini]

Originally posted by Nonsuch:

A lot of people do take Linus very seriously. I enjoy reading Slashdot forums where some young Linux kid posts about how microkernels suck and are slow, and then proceeds to get the sh!t flamed out of him for pages afterward.

Heck, I take him seriously because he's a pretty frickin smart guy!

[waffffffle]

Some interesting stuff that my prof said:

NT started as a microkernel, but they found it to be too slow so they started putting more and more stuff in the kernel, most importantly all of the display graphics stuff, thus making the interface fast. Could Apple put the entire graphics layer into the kernel? Is that a bad idea now that Darwin and Mach are opensource? I know it screws up that whole diagram that Apple shows with the different layers, but if the graphics layer was in the kernel I would think that Quartz as we know is would be much faster.


Also my prof makes us write feedback after each lecture and he writes back to everyone in one long email. I wrote to him: "Is the microkernal idea outdated? I use Mach (Mac OS X) and from my short experience it seems like a very good idea." He rephrased my question and answered it:

Is the microkernel idea outdated (MacOS X runs on top of Mach)?

There are probably still performance hits, and these might show up if
you write OS-intensive programs that really stress the boundaries.
However, that's usually not the kind of workload most people run on
Macs, so it's probably a non-issue.

[Nonsuch]

Originally posted by waffffffle:
Also my prof makes us write feedback after each lecture and he writes back to everyone in one long email.

What a cool idea! Wish my profs had done something like that.

[macmike42]

Originally posted by waffffffle:
Could Apple put the entire graphics layer into the kernel?

Yipes! Don't do that Apple! I like being able to kill the window server with EscapePod when it dies.

Besides, I think NT has the video drivers built into the kernel, not the window manager, which is the same as OS X (and most other modern OS's).

[benh57]

Originally posted by waffffffle:


Also my prof makes us write feedback after each lecture and he writes back to everyone in one long email. I wrote to him: "Is the microkernal idea outdated? I use Mach (Mac OS X) and from my short experience it seems like a very good idea." He rephrased my question and answered it:

Is the microkernel idea outdated (MacOS X runs on top of Mach)?

Dude, read the posts above. The Kernel that OS X runs on is **NOT** a microkernel *AT ALL*.

Period.

-B