Avie Tevanian Leaving Apple?

[Evangelion]

1. Gives apparent independent ammunition to the pro-Linux (or pro-monolithic) kernel camp, who are convinced Apple went in the wrong direction with XNU and Mach.

The test might be poor, but it's NOT poor because it shows OS X in bad light, while showing Linux in a good light. The reasons why it's a poor benchmark are in the technical details, and not in the fact that their results are not favourable to Apple. Had they done identical testing, but their results would have shown OS X beating Linux, would it then be a "good benchmark"? Nope, it would still be just as flawed as before.

That said, all this discussion about kernels in pointless. Linux has a good kernel, even though it's monolithic. I would MUCH rather have a good monolithic kernel than a bad microkernel. Saying that Linux sucks because it's monolithic just doesn't cut it. What would cut is telling the actual downsides of the kernel. And the fact that Linux is monolithic doesn't seem to be hurting anyone. It's still very stable and very portable.

It doesn't really matter what kind (monolithic, microkernel etc.) the kernel is. What matter is what the kernel can actually do. Should we be marvelling the superiority of some microkernel, even though it crashed every five minutes and had dismal performance? Should we call some monolithic kernel "obsolete", even though it was rock-solid, had excellent performance and was very portable? No and no. Theoretical merits of the kernel are irrelevant if the actual implementation sucks.

No, I'm not saying that OS X kernel is bad. Yes, it propably has it's share of problems in it. And so does Linux. And I'm pretty sure that those problems can be fixed.

 

[peharri]

The test might be poor, but it's NOT poor because it shows OS X in bad light,

You're misunderstanding me. I was responding to the question as to why the Anandtech article was causing a flame war. My two enumerated reasons weren't "Why the test is poor" but "Why such this poor test is causing a flamewar".

I pretty much agree with you, BTW.

 

[Evangelion]

You're misunderstanding me. I was responding to the question as to why the Anandtech article was causing a flame war. My two enumerated reasons weren't "Why the test is poor" but "Why such this poor test is causing a flamewar".

Ah, OK. Sorry for ranting

 

[dernhelm]

I don't see them doing it. We're talking about hundreds, possibly thousands, of dollars worth of programming time just to gain some fractional "efficiency" and reduced efforts involved in supporting third party tools and hardware. Why would they?

I hope you mean hundreds-of-thousands of dollars (more like millions) worth of programming time, not hundreds, possibly thousands. Hundreds of dollars is probably the donut budget for a week. Hmm - give up donuts this week and buy us binary level Linux compatibility? Yeah - you do that.

There's actually a lot of good reasons why Linux under the hood would be cool - none of them performance related. Mainly, porting of open source applications would be instantaeous - and OS/X would instantly become the most popular Linux distro.

As far as impact at Apple, OpenStep has been ported just about everywhere - including places without a Mach-based kernel. So most of Apple's software projects probably wouldn't notice the difference, since they are all built on top of OpenStep.

The main problem is that it would mean (most likely anyway) that current OS/X software would end up being "thunked" down into the Linux core, and would actually run slower until you received new binaries built directly on the new Linux core version of OS/X. And Apple cannot afford to have yet another code migration going on right now. So while it is cool to dream, don't hold your breath for this one.

 

[boncellis]

As long as they still have Ive, I'll be happy though. Ive's my hero!

Mine too, but Shhh! We'll be labeled forever if we say something like that!

 

[calculus]

Mine too, but Shhh! We'll be labeled forever if we say something like that!

Have to agree with you about Mr. Ive. For me he is the real Apple genius. All this technobabble above just turns me right off.

 

[Silencio]

This is non-news, period.

Avie busted his hump for 6 or so years turning OpenStep into Mac OS X and rapidly making it a mature product. He got bumped "up" to the position of Chief Propellerhead in 2003, where he was not involved in day-to-day decisions about Mac OS X at all. Since that time, his fellow ex-NeXTie, Bertrand Serlet, has held the reigns on software development at Apple and is pretty much continuing on the same course and philosophy as Tevanian has. Maybe their stance on file system metadata has softened a bit, which is a good thing...

Basically, this is like Fred Anderson leaving: the chain of succession in both cases has been completely seamless.

There are only two non-replaceable executives at Apple: Jonathan Ive and Steve Jobs.

 

[aegisdesign]

Thanks for that, I wasn't aware of those problems in Anand's explanation. However, the results remain the same. OS X is much slower than Linux for these operations, and I don't doubt that this is partly related to the half-monolithic/half-microkernel Frankenstein's monster that is the XNU kernel and the additional IPC overhead involved with microkernels.

Nope. It's due to the way MacOSX running on HFS+ does it's fsync() in MySQL. Linux and many other UNIX systems will flush data to the disk and then return immediately even if the disk hasn't actually written data. MacOSX waits for the disk to say it's actually written. If you've a badly written application like MySQL that frequently flushes data, you'll get bad performance.

If you want fast, use Linux and a big UPS power supply. If you want safe, use OSX. If you want both, use something other than MySQL.

This is pointed out by Apple's Dominic Giampaolo (of BeFS fame) in the MySQL docs at http://dev.mysql.com/doc/refman/4.1/en/news-4-1-9.html however Anand paid no attention to this and went off on some completely misguided discussion about threads, forks and kernels. This myth has now propagated to stupid proportions.

 

[aegisdesign]

NeXT/Apple merger was a brilliant move, the next brilliant move would be for Apple to buy Palm, me thinks -- an Apple Treo would be great.

Palm don't make software anymore though. That was sold to Access last year. That would leave just the hardware and I'm sure Apple can do better.

 

[aegisdesign]

I would really like to know what companies Rubenstein and Tevanian are going to and the reasons for ther moves.

Rubenstein is supposedly retiring.

 

[50548]

Palm don't make software anymore though. That was sold to Access last year. That would leave just the hardware and I'm sure Apple can do better.

I don't which rumor is older: Apple buying Palm, the PowerBook G5, or the iPhone...

Palm is all but dead...let's forget it, please...

 

[mdriftmeyer]

I'm saying that there are ZERO reason why Linux and OS X couldn't co-exist with each other. But you guys seem to want to make that as hard as possible.



You could say the same about OS X as well. For many cases Windows would be a better choice.



So is OS X. The key is to hide that complexity. Linux can be complex to the user. And it can be very simple as well.



Opinion, not a fact. When I started using OS X after using Linux for years, I found that OS X had "software base problems". All the apps that I used and loved in Linux were simply not there. And the apps that I did manage to find, usually cost money.



How so? On my Linux (Kubuntu) I have basically two means of installing apps:

a) I launch a package-manager. I select the app I want in the package-manager and click "install". The app is installed

b) I have installed Klik on my machine. That means that I can install apps by (*drum roll*).... klicking on a link in a website. That's it.

How exactly are those methods "cumbersome"? Or are you whining because it's different from OS X, and everyhting that is not like OS X is "cumbersome"? Or, IOW, you are stating your subjective opinion as fact.



No it isn't. you are stating your personal opinion as fact. Is your personal opinion worth more than my personal opinion?



Again: Opinion, not a fact. I HAVE seen lots and lots of "closed boxes" that run Linux, and they do it with zero issues. Windows would suck, since its full of spyware, viruses and other miscellianeous crap. OS X would be better, but it requires specific hardware to work. Linux has the advantage in that it would work on just about all hardware out there.



That's your choice. But just cut out the BS, OK? Dont like Linux? Fine, you have that choice. But please: don't spread bunch of personal opinions as facts. There really is no need for the hatred towards Linux. Linux did not steal your girlfriend, it did not scratch your car nor did it drink your beer. I honestly don't understand the hatred towards Linux. Success of Linux does NOT take anything away from Apple. Fact is that Linux could be Apple's and OS X's best friend in the computing-world. An OS that actually implements and follows open and documented standards? An OS that actually tries to work well with other OS'es? And still, all we ever hear from Mac-users is bitching and moaning, as far as Linux is concerned.

It just boggles the mind, really.

Being an OS X user/dev and Linux dev on Debian for the past 5 years if Linux wants to cozy up with OS X then it better start helping out GNUstep. The fact neither KDE nor GNOME have ObjC/Cocoa/GNUstep bindings into their tools is a problem. Apple doesn't need to add GTK+/Qt support for their commercial OS to be a success. It has better frameworks in Cocoa. Granted I'm biased having worked at both NeXT and Apple but it is clear that these C++ toolkits have been playing catch up to Cocoa for ten years and still won't match the elegance due to the obvious differences between ObjC and C++.

Now that ObjC++ is available for GNUstep I still expect no one to use it but the talented GNUstep devs who will leverage the many useful and well-written C++ libraries/frameworks/toolkits under GPL.

 

[mdriftmeyer]

Nope. It's due to the way MacOSX running on HFS+ does it's fsync() in MySQL. Linux and many other UNIX systems will flush data to the disk and then return immediately even if the disk hasn't actually written data. MacOSX waits for the disk to say it's actually written. If you've a badly written application like MySQL that frequently flushes data, you'll get bad performance.

If you want fast, use Linux and a big UPS power supply. If you want safe, use OSX. If you want both, use something other than MySQL.

This is pointed out by Apple's Dominic Giampaolo (of BeFS fame) in the MySQL docs at http://dev.mysql.com/doc/refman/4.1/en/news-4-1-9.html however Anand paid no attention to this and went off on some completely misguided discussion about threads, forks and kernels. This myth has now propagated to stupid proportions.

It's amazing we don't read about Oracle 10 or PostgreSQL 8 having these issues on OS X. Or Sybase or OpenbaseSQL for that matter.

 

[rayz]

It's amazing we don't read about Oracle 10 or PostgreSQL 8 having these issues on OS X. Or Sybase or OpenbaseSQL for that matter.

Not that amazing really. MySQL is probably the database that most folk will use on MacOSX because it's free.

I don't think there are that many serious Oracle10 deployments on OSX.

 

[rayz]

I don't which rumor is older: Apple buying Palm, the PowerBook G5, or the iPhone...

Palm is all but dead...let's forget it, please...

Didn't folk say the same thing about Apple when it was in a much worse position?

 

[Evangelion]

Being an OS X user/dev and Linux dev on Debian for the past 5 years if Linux wants to cozy up with OS X then it better start helping out GNUstep. The fact neither KDE nor GNOME have ObjC/Cocoa/GNUstep bindings into their tools is a problem.

Couldn't you just as well say that "Apple needs to start helping out GNOME/KDE"? Does Apple have support kdelibs and/or Qt in their tools for example? How about GTK+? No? Then why should Linux-folks pick up the slack, whereas Apple doesn't have to? "Because Apple-toolkits are better!". Well, I know bunch of KDE-developers that would disagree with you .

I'm not talking about sharing code or something here. I'm talking about co-existing in a way where one party does not actively seek to destroy the other.

Apple doesn't need to add GTK+/Qt support for their commercial OS to be a success.

And Linux-folks don't have to add support for Cocoa or others in order to be a success.

 

[peharri]

Not that amazing really. MySQL is probably the database that most folk will use on MacOSX because it's free.
OSX.

PostgreSQL is also free. It's generally considered superior to MySQL except in not being as easy to set up.

 

[janstett]

As you see, most modern OS use microkernels, for various reasons. I invite you to take a look at the BeOS development, QNX development (there are even some nuclear plants that works in an environment based in QNX) and other "modern" Operating Systems (I mean modern by using new concepts of kernel, not the 80s monolithical kernel) and you will know what i'm talking about.

Sometimes the theoretically superior idea (microkernel) ends up being inferior in real world performance.

Microsoft figured this out several operating systems ago. Dave Cutler designed NT to be a portable microkernel (remember, it has run on i386, Intel N 10, PowerPC, DEC Alpha, Mips, Itanium, and x64 platforms over its life). For Windows 2000, they bled pieces of the kernel and user space together to reduce these context switches. Precisely because the microkernel is a great idea, but the reality is that the OS then has to make a LOT of context switches in and out of the kernel to do every single little task. XP today benefits from this hybrid approach.

 

[janstett]

What other interests, I wonder--you're already at one of the top computer companies in the world...

This seems strange to me, I suspect certain code words like "mutual decision" and "amicable" might be involved.

Just a feeling.

I dunno, how exciting is it to be an Operating System guru in 2006? What new has really been done? It's a mature technological area now, with only small tweaks possible. The OS's now are pretty stable. Windows doesn't change much. Unix doesn't change much. Most embedded environments use some permutation of Unix.

I mean, you've got Kernighan and Pike from Bell Labs, who invented Unix. They finished their careers as researchers.

Take the early Microsoft guys -- Gordon Letwin did early Windows and OS/2 1.x, I haven't heard a peep out of him since he was forced out by Dave Cutler's arrival at Microsoft.

Dave Cutler designed VAX/VMS and then went to Microsoft to create NT. He left, and I haven't heard anything since.

Then take our older Mac OS gurus, the guy who left to start Be, etc.

There's not much exciting to do, being an OS guru, anymore. It's kind of like trying to invent a new kind of internal combustion engine -- you can improve small things but the design is fixed now.

 

[naturetrance]

This guy is cashing out (check AAPL insider stock sales) while Apple is on top. Smart guy.

I'm sure there are other reasons for his leaving though.