Pete Goodwin wrote:
> In this article:
> Lab Report: Windows XP Outperforms Earlier Versions
Consider the source. Have you EVER seen Microsoft publish a
benchmark or test result that didn't say whatever they wanted
it to say, something that would increase Microsoft's revenue
and market share?
Look at what they were evaluating:
<b>Faster startup performance: Windows XP is on average 34%
faster than Windows 2000 and 27% faster than Windows 98
Linux and UNIX can do this to by loading an application and
setting the "sticky bit". We can also preload the xrdb with
all of the app-defaults, it makes things much faster, but
Quite simply, Microsoft realized that RAM would be very cheap,
so they packed 98% of the code for Microsoft Office into DLLs
that are sucked into the system. Even if you only have 128
meg, a decent swap space will eliminate the time required to
search directories and "honestly" load the applications.
When all you need to do is dump everyhing from disk to RAM, and
you put all of you network stuff in parallel, you get a faster
load. It still takes a very long time. One of the complaints
with Windows 2000 is that it takes a very long time to load.
Of course, if you are trying to start 3rd party software, on a
with only 128 meg of RAM, and you developed it using Kylix,
to take MUCH longer to load. It's an antitrust violation, but
the huddled masses happy, and who would know?
<b>Better run-time performance: This measurement refers to the
speed at which Windows XP performs tasks while your computer is
running. Improvements in Windows XP runtime performance are
evident in application startup and time and resource
management. For example, average application startup on Windows
XP is 25% faster than Windows 98 SE and equivalent to Windows
Again, this was based on single-tasking benchmarkes of
well-known office applications. Yes, Microsoft has reorganized
the functions in their DLLs to eliminate "Tramp Swapping"
(swapping of unused code) by coding the methods and organizing
each method in a separate object so that the most frequently
needed code would always be together. It's a feature that UNIX
has had for 20 years. Microsoft opted not to do this
initially, because they wanted to keep everything "object
oriented and multithreaded". By becoming more like UNIX, they
improved the product -- interesting.
<b>Memory and Performance: In systems which include the
recommended memory requirement of 128 megabytes of RAM, Windows
XP is consistently superior to previous versions of
Again, the Windows 98SE and ME code was still very "object
oriented" and used the cache very ineffeciently. Microsoft
made it more like Linux/UNIX,
and things improved. Also note that the Windows 98/ME kernel
was trying to manage more with threads and doing more in the
application. Depending on which applications you are using for
the test, you will get different
Most of these optimizations were started with Windows 2000.
Again, this was done by Linux and X11 LONG before Microsoft
thought of it. Funny, isn't it, the more like UNIX and Linux
Windows becomes, the better it performs. Another 5 years and 3
"breakthrough releases" and they might even get it right.
<b>The lab compared the performance of nine desktop computer
systems. The computers were four of the most up-to-date,
high-end desktop systems; two mid-range desktop systems; a
high-end notebook system; one circa-2000 mid-range desktop
system; and one old desktop system. The computers were tested
running each version of the operating system at several memory
Keep in mind they had to go through nine different
configurations, and ran a number of different tests, to pick
the ONE configuration that have Microsoft the result it
wanted. It would be very interesting to see what the REST of
the configuration tests looked like.
> I see the following claims:
> "On average, Windows XP-based computers:
> * Score 36% higher than Windows 98 SE on Business Winstone 2001.
> * Score 77% higher than Windows 98 SE on Content Creation Winstone
> 2001. * Perform equivalent to the record-setting speed of Windows
> 2000 Professional, even with the addition of extensive new
> productivity features."
> Yet elsewhere in COLA, I see postings claiming the exact opposite.
> Which one should anyone believe? The Microsoft one (above) which is
> bound to say "it's faster!" or the COLA ones which are bound to say
> "it's slower!".
Neither, and both. The fact is that Microsoft has incorporated
many features from the UNIX and Linux playbooks to improve
Windows 2000 and Windows XP. The Windows 2000 kernel is much
more unix-like than it's NT predecessors. Unfortunately, you
have to have applications especially designed for Windows 2000
to get many of the performance and stability gains. The old
applications will run, but not as sharply as new ones using
MSMQ, and COM+.
> I think I prefer The Register's take on this:
> Is XP performance worse than Win2k, or just the same?
> "A pretty comprehensive set of benchmarks over at The Tech Report
> comes up with the possibly less than earth-shattering conclusion that
> WinXP is pretty much neck and neck with Win2k, from a performance point
> of view. XP seems to be slightly better than Win2k when it comes to
> office productivity, but is pretty well tied with it on graphics and
> gaming, in some cases even marginally losing."
No surprise here. When you put the nearly all of "Office" into
DLLs that are preloaded at boot-time, and those DLLs are
optimized for the L1 and L2 cache, your "Office" performance
will be faster.
When you are running 3rd party software, especially things like
you have to compete with Microsoft for RAM, L2, and L1 cache.
You keep paging rows in, and the OS keeps bumping you rows out.
> On my home machine, Windows XP boots faster than Windows ME.
Again, not really a suprise. Linux prefetches entire tracks at
a time, which reduces rotational latency significantly.
Windows is now smart enough to do this when loading DLLs.
Again, Microsoft made their system more like Linux and UNIX,
and ended up with a better system.
> I can't
> say anything about the speed of applications, as I've not compared them
> as yet. The slowest system to boot is Linux Mandrake; it's also the
> slowest to login.
Actually, there is a reason for this. While Microsoft XP only
supports the latest hardware, and their "blessed" peripheral
busses (USB, PCI, FireWire), they don't support VLB, older
technologies, or non PnP devices. If you have an old 3C509
card, jumpered for IRQ 7, Windows XP will just stomp all over
it, Linux will detect it, and protect it.
Linux could take some of the shortcuts that Microsoft takes,
but then you wouldn't get the best DHCP host, the best DNS, the
best usage of IRQs, and the best use of memory. Linux loads
lots of modules during boot time that Windows only probes for
during the initial set-up. With Windows XP,
it costs you one of your 3 activiation keys each time you
When I run Linux on my laptop, and jump from customer site, to
to cable modem, to office, I can easily have 4 different
hardware configurations. Linux tunes and optimizes my system
each time. In addition, I add USB devices, IEEE1394 devices,
and PCMCIA devices
in different configurations as well. Windows wants to know
about every peripheral, and tunes as if all of it were running
- the token ring card driver conflicts with the Ethernet card
driver, but I'm only using one or the other.
> Perhaps the answer to this question is: "Who cares?"
It's true. If you want a computer that's easy to learn,
make you think, and tells you how to run your business, and
life, exactly the same way it tells your competitors, then
is your system.
If you are willing to do a little bit of self-directed
learning, using the resources included, you can save a bit of
money, get better solutions to your problems, and gain a
competitive advantage. You can have more choices, and you can
"have it your way".
It really isn't just about which machine can recalculate a
spreadsheet in the fewest number of microseconds, it's really
about whether you want to spend a premioum price for "Off the
Rack" or "Off the Lot", or have your system "Custom fit" to
your needs and preferences.
It is much the same as the choice people made in 1994-5. Which
would they rather have, 5 different services, each charging
$20/month, to provide proprietary content from a limited number
of providers who paid a premium (as high as 80% of all revenue
raised) for "per hour" services, or a dial-up connection to a
TCP/IP service, with access to several million sites, each
providing quite a bit of free content, along with advertizing
that was closely related to the subjects discussed in the
Today, almost everybody has chosen choice number 2. Even if
they choose to join AOL or MSN, they EXPECT to be able to
access any site on the Internet, and be able to communicate
with anyone, via e-mail, chat, or web. How quickly would MSN's
loyal customers, most of whom just joined MSN for the $400
rebate, would drop MSN like a hot potato if Microsoft tried to
restrict access to "Microsoft Approved Sites Only". Suddenly,
they would be dialing through another service, or a local ISP.
The same is true of broadband media as well.
The Linux/XP choice is almost identical. We've seen how
Microsoft treats it's competitors. We've seen non-microsoft
products become dust-collectors, and we've seen archives become
inaccessible because we can't run the Windows 3.1 or MS-DOS
application that could read them, on Windows 9x or 2k. We've
read more »