10 Things Apple Did To Make Mac OS X Faster

bariswheel writes "This kernelthread article seeks to investigate further to the inner core of OS X and the improvements therein. The subtopics are the following: BootCache, Kernel Extensions Cache, Hot File Clustering, Working Set Detection, On-the-fly Defragmentation, Prebinding, Helping Developers Create Code Faster, Helping Developers Create Faster Code, Journaling in HFS Plus, and Instant-on."

I love OS X

OS X is the only OS I"ve ever installed that subsequent versions speed up my older computers. Amazing... I'm waiting for an Apple Intel Tower and I'll retire my G4 Tower.

Damn ADC interface.. what am i to do with this big ass cinema display?!?!!?

Re:I love OS X

Clearly you've never installed Windows 2000 over Windows ME.

Re:I love OS X

Um.. ANYTHING installed over Windows ME is an improvement. Hell, Going backwards would be an improvement.

Re:I love OS X

Indeed. Windows ME was a crime against humanity.

Re:I love OS X

I installed Windows ME when it came out.

It came off my machine after a month, and I went back to Win98 SE.

Yes, it WAS that bad.

Re:I love OS X

"1. faster booting"

Yep, comes in handy when the OS can't handle a day or two of uptime. Windows 2000 was so much more stable, and didn't take all that long to boot. Longer then ME, yes, but I bet you wasted more time watching the boot screen then 2000 users did.

"2. disk scan ran inside windows and was a million times faster"

Except that ME wasn't smart enough to multitask when scanning a disk. So that frequent bootup disk scan you saw was always interrupted several times when it tried to start, and if some bootup process accessed the disk say every minute or two, it would never finish. I think the majority of ME users just cancelled that any time it popped up. Of course those of us who skipped ME and went from 98 to 2000 started enjoying journaled filesystems and had no need for the disk scan to run inside windows.

"3. native .zip support"

Zip support that is horribly implemented. Lets walk you through a multipart wizard to extract this file, or present it as an explorer window that lets you run things directly out of, but causes most programs to freak out when you try this. I still don't use the built in Zip support on XP even though it has been slightly improved. Running things inside a Zip directly is as bad as compressing the hard drive for more space.

ME sucked. It was simply a quick release from Microsoft for the consumer market to get something new out, since all the "consumer friendly" features didn't make it into NT 5, err, I mean Windows 2000. For MS to go completly backwards and ship another archaic 16/32 bit mess of DOS based code after Windows 2000 was just silly. I feel pitty on anyone who actually paid for a copy of ME.

Linux

Linux gets faster too.

Kernel 2.4 to 2.6 was a pretty big jump in speed. I just upgraded to the latest KDE and a bunch of other updates, and got another performance jump. Once they shake the bugs out of the Radeon drivers for X.org, I'll get accelerated X, and another big speed boost.

In fact, of the major OSs, it's pretty much only Windows that keeps getting slower.

Re:Linux

Kernel 2.4 to 2.6 was a pretty big jump in speed.

That's true, but don't expect another jump of that relative magnitude anytime soon. The 2.6 introduced a new scheduling algorithm that boosted speed and concurrency significantly. When your scheduler goes from an O(n) to an O(1) algorithm, implements CPU affinity, and eliminates lock contention for the run queue, the speed boost is significant. I guess there could be filesystem improvements or paging improvements in the pipeline that could provide significant speed boosts, but I kind of doubt they would be as critical as that brilliant new scheduling algorithm.

OTOH the inter-version speed boosts in OS X have been due to more subtle tweakage, except perhaps for speed boosts related to launchd, and have been more incremental in nature than the anomalous 2.4-2.6 improvement.

I guess my point is that the 2.4-2.6 improvement is more of a leap than it is a trend, where OS X's improvements have been less revolutionary and more evolutionary. I hope Linux continues to improve in performance, but it's very possibly going to suffer from bloat down the road that could offset some performance improvements. It's unrealistic to expect the performance improvements to continue along the lines of 2.4-2.6, in any case. OS X is still lagging in performance, so it's even more imperative that it continue its trend. Hopefully the researchers at Apple will soon find a revolutionary improvement on the order of the 2.6 scheduler to catch up a bit.

Re:I love OS X

Damn ADC interface.. what am i to do with this big ass cinema display?!?!!?

As you may or may not be aware, the ADC connection provides a DVI signal, USB port, AND power. The display has no power pack, and gets its juice from the computer. If you have only a DVI port, you will require a rather large adapter. It's not so much an adapter as it is a "power injector" that injects power into the cable whilst converting it from DVI+USB to ADC. This takes the form of what looks like a very large white power brick from a powerbook.

They are unfortunately rather expensive. ($150?) You can get them from Apple, or from Dr Bott.

The other answer is of course to find a graphics artist or developer that does not already have a second display, and sell it to them. Odds are very hight that if you bring the display over and let them "test drive" it for even five minutes they'll buy it immediately.

Re:I love OS X

And if you bitch about having to buy an adapter to drive your Cinema with your new Mac, they may give you a 99 dollar discount right over the phone. They did for me when I bought my Quad.

Dupe several years later?

Check this: http://developers.slashdot.org/article.pl?sid=04/0 6/03/130214 [slashdot.org]

Obvious Dupe

The website even has a link to the old slashdot story: http://developers.slashdot.org/developers/04/06/03 /130214.shtml [slashdot.org]

Pointless Effects

If Apple is going to bother optimizing other stuff on the OS, they should at least give you a way to turn off some of the extras when it comes to the GUI.

I don't need high resoution icons, drop shadows, dragging window effects, minimize effects...etc. In windows land, you can turn most of these eyecandy effects off and performance is greatly improved. You'd think that Apple would have considered this when releasing a computer with 256mb of ram on the base model (G4 mac mini). I love the computer, but it is SLOW.

Re:Pointless Effects

I don't need high resoution icons

Those you can turn off. Just set the Finder to use 32x32 pixel icons. icns resources generally contain several versions of an icon, 128x128, 48x48, 32x32, and 16x16 pixels. If you use one of the small versions then the system won't waste time scaling the icon, or memory holding a big bitmap. I doubt you'll see much gain though.

But it's not in Apple's interest to let you turn off too much of the eye-candy. They want Mac OS to have its distinct look, and they are are in the business of trying to sell you newer hardware.

Re:Pointless Effects

Unlike Windows, OS X is composited on the video hardware, and the effort to produce most of those visual effects is done by the GPU, hardware that would otherwise be idle. Turning them off wouldn't give you any speed gains on the CPU, from what I understand.

Re:Pointless Effects

In Windows land, the desktop eye-candy isn't hardware accelerated. Turning off a lot of the OSX eye-candy would only serve to idle the graphics hardware rather than making the computer respond any faster.

Hopefully, Microsoft's Aero will prove this point.

Re:Pointless Effects

those GUI 'extras' are not what is making a 256MB G4 slow. Rather, it would be the fact that the machine is going to be constantly swapping out to disk. Get more RAM.

Re:Pointless Effects

Not knowing much about you, the only thing that comes to mind is to stop externalizing blame. I don't have much to go on here, though.

I'm not a SO guru, but...

what about Linux? Could it obtain benefits implementing some of the improvements made into MacOS X? I've heard about BeOS and the incredible perfomance due to multithreading, it's very dificult to adapt an BeOS kernel to the Linux features (multiuser, drivers...) maintaining the perfomance?

Panther to Tiger?

I updated from Panther(10.3) to Tiger(10.4) and my machine seemed slower. I decided to do a fresh install, and things improved, as always the fresh install is better than an update.

I still think that Panther was running a bit faster tahn Tiger, maybe it is the widgets..........
silly widgets!

This was all done on a PowerBook G4(TiBook).

Re:Panther to Tiger?

It's not just you, I've heard many reports of Tiger being slower on older machines. Because of that, I'm staying with 10.3 on my Pismo until it is replaced as it runs acceptably fast in nearly every situation (I don't do video or gaming or any other CPU/GPU intensive stuff). I don't miss Widgets or any of the other new eye-candy type stuff in 10.4. BTW, the replacement for Pismo will be in the form of the second-gen Macbook Pro with a Merom-based core duo being released in Q3 (fingers X'ed), and will hopefully last 6 years like my Pismo has.

Re:Panther to Tiger?

I have two slower Macs, A G3 500 MHz iBook running 10.4.5 and a Blue and White G3 400 MHz running 10.3.9. The iBook is a bit faster for everyday tasks and that hasnt always been the case (the tower has a faster bus, faster graphics card, faster disk etc...). One thing you MUST do on older macs running Tiger OR Panther is upgade your RAM to a reasonable level, which usually means maxing it out. Even then I had to turn off dashboard on the iBook (I dont really use it on any of my macs anyway).

Both machines are still great for general desktop work and light development. I bumped the iBook to Tiger (OSX) to get Tiger (Java), and I havent really bothered to upgrade the tower because I havent had the time and its not a pressing concern for me. But given the results on the iBook, I dont expect a performance hit when I do upgrade.

Re:Panther to Tiger?

I still think that Panther was running a bit faster tahn Tiger, maybe it is the widgets..........
silly widgets!

No, it's spotlight. My iBook would thrash like crazy until I disabled spotlight. Of course now I can't search at all.

Apple should've made spotlight optional.

Re:Panther to Tiger?

Get EasyFind from http://www.devon-technologies.com/products/freewar e/applications.php [devon-technologies.com]. It works without Spotlight.

Ten things they should fix

somebody made a list about ten things that don't work as well as they should (and as a mac admin I agree) : Ten More Things I Hate About Mac OS X [informit.com]

rest of my reply

since I screwed up, here's the rest:

I'm not sure I agree with all or even most of his points of contention.

In issue 1 for example he complains that each open/save dialouge starts out the exact same way and then goes on to complain further in the article that the OS isn't always consistant. It's consistant for each dialouge to remain the same size by default until the user specifies a change. Furthermore since the size of the dialouge can be set per application, that size would need to be specified by the application making having a universal override obnoxious.

In his 2nd point he's descirbes a senario which is at best extremely uncommon and then describes a process which is obnoxious and complicated when it's easier for most people to either have an automator script to open specific things they want or even better for his senario and automator script which asks where he is and then opens the appropriate applications. A simple applescript for the applications one doesn't need all the time with a prompt at the begining to ask whether to launch the remaining apps and then placing that script in the login items folder seems more useful and less annoying than check boxes to enable and disable each item that you must do before loging out the previous time.

point 3 he's correct on

point 4 he's correct on the disapearing sidebar but on the issue of double clicking the boarder, it's a rather difficult task to accomplish accidently so I am sure anyone doing it would notice the dimple before and after.

point 5 he's moving away from his consistancy argument again. With the column view you set the size of the columns and the number of columns, and if you chose to physicaly change the display you can. What he's suggesting is a display system which dynamicaly changes size to fit the content of the display which while it could be benneficial to some people seems overly complicated and a major violation of the consistancy guideline. It's concieveable to see a situation there where all of a sudden you would go from having 4 collumns displayed to having 2 or 1 because you have one file in the display such as "com.apple.Components2.LocalCache.QuickTimeCompone nts" which now expands their one collumn to occupy most of the window.

point 6 he's correct on

point 7 he's got a point but at the same time, with the addition of the PDF abilities and the fact that faxing IS handled with PDFs it does make sense to put it under the PDF button. In the end I don't find it much more of an abstraction than his recomendation to make it an availible printer.

point 8 I can see a method to the madness in that if the next set of startup items require the server, it's important for you to know that the server is not availible BEFORE those apps launch and fail. There may be a better way, but I don't agree that it's a failing.

in point 9 the views update for the column view I think is a good thing. While it's not 100% consistant, in this case it would be irritating for a directory I'm working with to rename and then immediately move out of my working view until I indicate being done with the directory either by being idle or moving to a new object.

The size information I would assume is an updating routine thats scheduled rather than called.

in point 10 if he cant see a situation where a user might unknowingly or mistakenly change their file extention then he needs to think harder. The checkbox would be nice though but it's also nitpicking at this point. It's a potentialy destructive action, and a user should be reminded to think before they do it. Being able to permanently dismiss such reminders is what gives viruses and other malicious programs a better chance of succeeding.

G3

a while back, when OSX.2 had recently came out, and I was still running on an iMac G3 (with 256mb of RAM), I got OSX. it ran much smoother than OS9 did.

Ironic?

Consider the following a sampling of such optimizations, in no particular order

I'm somewhat concerned that an optimisation geek did not order his data set.

Whats up with the ABI change?

It used to be that OSX had a brain-dead ABI that resulted in not all of the PPC registers being used 'properly', in order to maintain a 68k 'compatability' mode ..

Has this been changed? Are all the registers of the PPC being used properly now? Is the PC register actually being used as a program counter, rather than one of the generic 32-bit registers?

On-the-fly Defragmentation

I know that NetWare (at least as far back as 5.1) used a different method to avoid the need for defragmentation, which basically just allowed the disk to read the sectors based on the physical location on the disk rather than the order they were needed. The thing was, without defrag tools, you couldn't even check to see if fragmentation was a problem. The same problem exists with *nix filesystems right now. Everyone says you don't need to defrag, but there's no easy way for the average admin to verify this.

The HFS plus approach seems like a good idea, but I'm wondering if there is a performance cost, both in CPU cycles and drive wear and tear. It also looks to me like the system could be defragging files that are already contiguous, but I may be wrong. Given that modern journaling filesystems (supposedly) are not likely to become fragmented in the first place, is this feature worth it?

I've always loved this about Apple....

While Windows gradually slows down with Service Packs, and completely grinds to a hault on all but fairly new systems with new versions, OS X actually manages to speed up due to developer optomization. Good strategy...

journalling...

Journalling makes files accesses slower. The only time it speeds anything is in the error case of unsafe reboots (crashes). And I get those so rarely, I'm sure that I come out way behind on performance due to journalling.

What is a defragmentation? ;-)

Hi, I use mac stuff since may 2005, I sold my boring/noisy/energy hungry/ugly... 2.4Gh P4 for a mac mini witch I sold last month for a core duo iMac. Sine then, I never had to defrag my disk or do anything else that Windows still needs not to sink. Since Windows can now run, with some work on your part, on a mac, why not then consider buying a mac instead of a regular Windows machine? After all, VISTA will probably not ship until 2007 or later. To this date, VISTA has been announced for 2003, 2005, 2006 and now 2007!!! ans OSX is there NOW "Happy macking !"

one big thing they may have missed....

Doesn't Apple use gcc?

I know gcc itself improved a very great deal over the same time period, and I have always assumed that the speed gains were (largely? mostly?) due to that, rather than wondrous new algorithms on Apple's part.

Linux and KDE sped up a lot too, over the same timeframe.

Apple and GCC

Apple has been contibuting to GCC [apple.com] too you know. Objective C support, PowerPC optimizations, etc (scroll down to optimizations). Another advantage of OSS. The improvements on their hardware were due to their own efforts, and much more radical than the increases to x86 Linux.

Unfortunately, on the Intel side, Apple is going with the Intel compiler, probably because it's faster than GCC Intel. No OSS. But maybe Apple doesn't need to contribute to that because Intel will keep doing good work.

Undocumented Number 11

QUICKSILVER

Get it
Use it
Good

( P.S. Caps Lock would have been autopilot for COOL, but the lameness filter caught me :( )

"Dave: Is Dead == Everybody"

Helping Developers Create Code Faster,
Helping Developers Create Faster Code

I can think of a few other useful permutations:

Helping Create Code Developers Faster
Helping Create Faster Code Developers
Helping Code Create Developers Faster
Helping Code Create Faster Developers
Helping Faster Developers Create Code
Helping Faster Code Developers Create
Helping Faster Code Create Developers
Developers Helping Code Create Faster
Developers Helping Create Faster Code
Developers Helping Code Create Faster
Developers Helping Faster Code Create
Developers Create Helping Code Faster
Developers Create Faster Helping Code
Create Helping Code Developers Faster
Create Developers Helping Faster Code
Create Code Helping Developers Faster
Create Code Helping Faster Developers
Create Code Faster, Helping Developers
Create Faster, Helping Developers Code
Create Faster Developers, Helping Code
Create Faster Code, Helping Developers
Code Helping Developers Create Faster
Code Helping Create Developers Faster
Code Helping Create Faster Developers
Code Helping Faster Developers Create
Code Developers Helping Create Faster
Code Developers Create Faster Helping
Code-Faster Developers Helping Create
Faster-Helping Developers Create Code
Faster-Helping Code Create Developers
Faster Developers Helping Create Code
Faster Developers Helping Code Create
Faster Developers Create Helping Code
Faster Code Helping Developers Create
Faster Code Helping Create Developers
Faster Code Developers Helping Create

Choose a research topic! Lucrative grants to be won! (Topics involving procreation by/of developers expected to go quickly.)

Does this fix OSX's bad Apache & Database

performance under load? As reported here: http://www.anandtech.com/mac/showdoc.aspx?i=2520 [anandtech.com] OSX has to this time had severe performance problems under load for Web applications, does it look like the improvements in this article will help?

Hot File Clustering

Does locating so many files that are accessed so often in one part of the disk cause that section to fail more quickly than if all of the "hot" files were to be left where they were? I'm wondering what this does to the life of mechanical drives.

Where I've found OS X to be slow

Network transfers. We chose Dell 1425s running Fedora instead of XServes running 10.4 largely because our benchmarks found the XServes had a huge network latency. It took us almost ten times longer to emit packets in some cases.

Amusing comment about "sleep"

I found the comment about "instant-on" sleep to be amusing.

Yes, we're fully aware that Apple systems can shut down everything execept the components necessary to refresh the DRAM.

The author of the article, apparently, has never used a PC notebook or desktop. Practically every well-behaved system made in the past 5 years, from the $150 eMachines desktops to my generic Compal notebook, supports the ACPI S3 state, which does exactly what Apple's "sleep" mode.

What's really slick about Windows is that the system can wake from S3 suspend and hibernate itself after a certain period of time. My system is set for 6 hours, which means that I don't have to wait for the system to restore during the day, but if I leave my system overnight or longer, I don't have to worry about suspend draining my battery (approx. 20% per day). I can even have different settings if the system is plugged in.

Lack of hibernate hurts

Lack of hibernation in Mac OS, in my opinion, hurts. For those who don't know, "Hibernation" is the term Microsoft uses for a state in which all of the contents of memory are saved to the hard drive and the system completely shuts down. When the system is booted up, that cache is read from the hard drive (and is almost always much faster than a full-on boot). Considering all the things that could go wrong, it works excedingly well. It's sort of like a better sleep, hence the name.

As to why it's needed: battery life. I can hibernate my Dell, unplug it for a business trip and it's still got the same juice a day or two later when I turn it on. When I do the same with my Powerbook G4, the battery often dies while it's asleep.

I'm hoping it's one of the things they add for 10.5.

Re:Lack of hibernate hurts

I don't know why people try to defend Apple on this particular design decision. There's absolutely no reason why hibernation shouldn't be included in OS X.

It could be that it's because hiberation actually does exist in Mac OS X. It's just not a well known fact. OS X 10.4's "Safe Sleep" (Google cache [72.14.203.104]) saves the active memory to disk when a Mac [laptop] goes to sleep...lest the power get interrupted. If one is so inclined, they can activate it, and even choose to use it by default. I've enabled it on my Mini, and it definitely works.

However, if you're not a Mac user, you may not appreciate how good the normal "Sleep" mode is. Unlike Windows, a Mac which has been put to sleep will resume almost immediately, and be instantly usable. My iBook can stay 'asleep' in my briefcase for ages, with very little battery consumption, and as soon as I open the lid, I am good to go. This impresses me more than words can say.

Journaling?

...Journaling in HFS Plus...

That's odd, I assumed Journaling gave a slight performance hit... unless you kernel panic etc. and your recovery has it's ass covered.

And then there's the fact that I just today figured out how to prevent chronic fatal drive thrashing cycles on my sidekick iMac G3 (running Tiger): format the drive HFS non-journaled.

Re:What about OSes with GNOME?

You don't get out much, do you? GNOME 2.14 is supposed to be extremely fast in comparison to previous releases, which were also faster than their predecessors.

Uhhh...I'm guessing if anyone's not getting out much....well, nevermind. If you can't say something nice don't say it at all.

Re:What about OSes with GNOME?

Gnome is maybe not the best example. I'd rather talk about lightweight window managers like blackbox fluxbox and openbox. I tried using Gnome 2.10 on FreeBSD 6.0 on a 133 MHz Pentium I laptop, it's as slow as a zombie turtle, as blackbox runs fast.

Re:What about OSes with GNOME?

You don't have to be a "GNU/Hippie" to use Linux, and there are plenty of reasons to do so, as well, not the least of which is that it's free and it'll run on that old P166 you bought over a decade ago. The "GNU/Hippies" you speak of are largely the guys who spend all day tweaking this and that to make sure the next release of your operating system is secure, productive, and pleasing to the eye, which you might notice Linux is becoming more and more, especially with user-oriented flavours like Ubuntu. The main difference is, the guys at Apple get paid for what they do, and the guys who contribute to Gnome do not. As such, Apple is a little further ahead, especially since their UI is more closely integrated into the core of the OS than Linux' is (and they don't have to contend with different flavours of hardware). Anyway, in closing, flamebait.