Performance Improvements
The last significant group of core improvements to be found in Vista is a collection of several changes Microsoft has made to various subsystems in order to boost overall system performance. Unfortunately, because Vista is still in a beta state where performance optimizations have yet to be turned on, not to mention the lack of diagnostic programs to test these features, we won't be able to bring you any meaningful performance numbers until the release version of Vista ships. However, we can talk some about the most important of these features.
Since XP was launched 5 years ago, broadband internet access has come to be the dominant connection method in much of the world. While XP had some targeted improvements to increase performance on these connections, Microsoft is taking it one step further in Vista with a heavily modified TCP/IP stack. The primary changes will be that the stack is now capable of a greater level of self-optimization by automatically adjusting some settings to maximize performance, and some overhead has been removed to better reach the maximum theoretical performance of high-speed LANs (GigE and higher). The new stack also has support for both IPv4 and IPv6 (instead of requiring 2 separate stacks as is currently the case), but the amount of hardware support to go with this is still fairly low.
The I/O subsystem has also been given a major overhaul, and like the forthcoming changes with DirectX 10, the focus is on better ways to schedule and execute I/O tasks. I/O tasks can now be given a priority level, so that unimportant tasks such as virus-scanners can be given a lower disk priority that will prevent them from being so detrimental to system performance.
The I/O system will now also be able to handle Microsoft's SuperFetch technology, which promises to improve memory usage by creating a third tier of memory pages between those important enough to be in RAM and those relegated to being in a pagefile. The new tier will be a class of pages that are important enough that they need to be accessed often, but are small pages that don't take much time to transfer; these pages can be stored in connected flash-memory devices which offer latencies measured in nanoseconds (a couple orders of magnitude faster than a hard drive page file), and in turn free up the main memory for more important tasks. It's an interesting idea that we hope will give some real benefit in the release version of Vista, and is likely to become more important in the near future as hard drives start to be released with integrated flash memory buffers.
Lastly, the kernel itself has been slightly overhauled, implementing new methods of memory and task management. Microsoft's pitch here is that it will be better optimized for dual-core systems, but since the NT kernel has always had a good level of SMP support for multiple processors, we're a bit skeptical on just what kind of performance improvements can be found here.
The last significant group of core improvements to be found in Vista is a collection of several changes Microsoft has made to various subsystems in order to boost overall system performance. Unfortunately, because Vista is still in a beta state where performance optimizations have yet to be turned on, not to mention the lack of diagnostic programs to test these features, we won't be able to bring you any meaningful performance numbers until the release version of Vista ships. However, we can talk some about the most important of these features.
Since XP was launched 5 years ago, broadband internet access has come to be the dominant connection method in much of the world. While XP had some targeted improvements to increase performance on these connections, Microsoft is taking it one step further in Vista with a heavily modified TCP/IP stack. The primary changes will be that the stack is now capable of a greater level of self-optimization by automatically adjusting some settings to maximize performance, and some overhead has been removed to better reach the maximum theoretical performance of high-speed LANs (GigE and higher). The new stack also has support for both IPv4 and IPv6 (instead of requiring 2 separate stacks as is currently the case), but the amount of hardware support to go with this is still fairly low.
The I/O subsystem has also been given a major overhaul, and like the forthcoming changes with DirectX 10, the focus is on better ways to schedule and execute I/O tasks. I/O tasks can now be given a priority level, so that unimportant tasks such as virus-scanners can be given a lower disk priority that will prevent them from being so detrimental to system performance.
The I/O system will now also be able to handle Microsoft's SuperFetch technology, which promises to improve memory usage by creating a third tier of memory pages between those important enough to be in RAM and those relegated to being in a pagefile. The new tier will be a class of pages that are important enough that they need to be accessed often, but are small pages that don't take much time to transfer; these pages can be stored in connected flash-memory devices which offer latencies measured in nanoseconds (a couple orders of magnitude faster than a hard drive page file), and in turn free up the main memory for more important tasks. It's an interesting idea that we hope will give some real benefit in the release version of Vista, and is likely to become more important in the near future as hard drives start to be released with integrated flash memory buffers.
Lastly, the kernel itself has been slightly overhauled, implementing new methods of memory and task management. Microsoft's pitch here is that it will be better optimized for dual-core systems, but since the NT kernel has always had a good level of SMP support for multiple processors, we're a bit skeptical on just what kind of performance improvements can be found here.
Facebook
Twitter
RSS
75 Comments
View All Comments
Pirks - Friday, June 16, 2006 - link
Excuse me, what? How about this then: http://www.macworld.com/2006/05/reviews/osxfirewal...">http://www.macworld.com/2006/05/reviews/osxfirewal..."The emphasis is on incoming. As it ships from Apple, the firewall does not monitor traffic that may be originating from your own computer. If your Mac gets possessed by a malware application that then attempts to attack or infect other computers via your Internet connection (a not-uncommon trick), OS X’s firewall won’t, by default, pay any attention. And, there’s no way to change this default setting from your System Preferences. To force the firewall to monitor outbound traffic, you must use Terminal’s command-line interface."
See - IT CAN monitor and block outbound traffic, contrary to what you say. It's just a matter of configuring it properly. You should at least correct your article and stop saying OSX ipfw CAN'T track outbound connections. You can say this: it's SET UP not to monitor outbound connections BY DEFAULT but anyone can CONFIGURE it to monitor outbound connections either through third party GUI like Flying Buttress or via command line. Then you won't look like a liar to any Mac guy who cares to read your review.
Ryan Smith - Friday, June 16, 2006 - link
I see your point, but I believe there's nothing in the article that needs changing. Tiger's firewall can't block outbound connections without having to drop to the terminal to muck with IPFW, I do not classify that as an ability any more than I classify Vista x64 as being amateur driver programmer friendly(since you need to drop to the terminal to turn off the x64 integrity check). When a version of Mac OS X ships with a proper GUI for controlling outbound firewalling(as is the Apple way), then it will be capable by a reasonable definition. Right now it's nothing more than a quirk that results from using the BSD base.Pirks - Friday, June 16, 2006 - link
Excellent point! So, when (and if) Mac OS X will see its share of virii and malware, THEN Apple will incorporate outbound connection settings in OS X GUI - right now it's not needed by Mac users, and the rare exceptions are easily treated with third party apps and command line.OK, got your point, agreed, issue closed. Thanks :)
bjtags - Friday, June 16, 2006 - link
Vista x64I have been pounding on it for 4 days never crash or even farted once!!!
Have all HalfLife 2 and CS running Just Great!!!
Had at one time 10 IE windows open, MediaPlayer, Steam updating, download driver,
updating windows drivers, and 3 folder explorer windows open, and tranfering
4gig movie to HD!!!
Still ran fine... I do have AMD 4800 x2 with 2gigs...
Poser - Friday, June 16, 2006 - link
Two questions:1. What's the ship date for Vista supposed to be? Q4 of 2006?
2. I seem to remember that speech recognition would be included and integrated with Vista. Is it considered too much of a niche toy to even mention, not considered to be part of the OS, or am I just plain wrong about it's inclusion?
It was a extremely well written article. Very nice job.
Ryan Smith - Friday, June 16, 2006 - link
1. Expected completion is Q4 with some business customers getting access to the final version at that time. It won't be released to the public until 2007 however.2. You're right, speech recognition is included. You're also right in that given the amount of stuff we had to cover in one article it was too much of a niche; voice recognition so far is still too immature to replace typing.
ashay - Friday, June 16, 2006 - link
"Dogfooding" is when a company uses their own new product (not necessarily beta) for internal use.(maybe even in critical production systems).Term comes from "eat your own dog-food". Meaning if you're a dog food maker, the CEO and execs eat the stuff. If they like it they dogs hopefully will.
http://en.wikipedia.org/wiki/Eat_one%27s_own_dog_f...">Wikipedia link
fishbits - Friday, June 16, 2006 - link
Yes, I know it's still beta, we'll see. The UAC and signed drivers schemes sound like they'll be flops right out of the gate. Average user will quickly realize he can't install or use anything until he adopts a "just click 'Yes'" attitude, which will reward him with a functioning device/running program. I've lost count of how many drivers I've installed under XP that were for name-brand devices, yet didn't have the official seal of approval on them. Again, get trained to "just click 'Yes'" in order to be able to do anything useful. Without better information given to the user at this decision point, all the scheme does is add a few mouse-clicks and no security. Like when you install a program and your security suite gives a "helpful" warning like "INeedToRun.exe is trying to access feccflm.dll ... no recommendation."As expected, it looks like the productivity gains of GPU-acceleration were immediately swallowed up by GUI overhead. Whee! "The users can solve this through future hardware upgrades." Gotcha. For what it's worth, the gadgets/widgets look needlessly large and ugly, especially for simply displaying things like time, cpu temp/usage. Then it sounds like we're going to have resource-hungry programs getting starved because of GPU sharing, or will have an arms-race of workarounds to get their hands on the power they think they need.
Ah well, I've got to move to 64-bit for RAM purposes relatively soon. Think I'll wait a year or two after Vista 64 to let it get stable, faster, and better supported. Then hopefully the programs I'll need to upgrade can be purchased along the lines of a normal upgrade cycle. Games I'm actually not as worried about, as I expect XP/DX9 support to continue for a decent bit and will retain an XP box and install Vista on a brand new one when the time comes.
shamgar03 - Friday, June 16, 2006 - link
I really hope that will mean for BETTER GPU performance not worse. I would really just like to be able to boot into a game only environment where you have something like a grub interface to pick games and it only loads the needed stuff for the game.darkdemyze - Friday, June 16, 2006 - link
beta implies "still in developement". chances are very high performance will see an increase by the time of release. I agree with your seconds statement though.