Slashdot Mirror
PC World Tests Final Version of Vista SP1
Mac writes "PC World ran the final version of Windows Vista SP1 through a first set of tests last night. Here's the bottom line: 'File copying, one of the main performance-related complaints from Vista users, was significantly faster. But other tests showed little improvement and, in two tests, our experience was actually a little better without the service pack installed than with it.'"
"The Windows Vista SP1 install process clears the user-specific data that is used by Windows to optimize performance, which may make the system feel less responsive immediately after install. As the customer uses their SP1 PC, the system will be retrained over the course of a few hours or days and will return to the previous level of responsiveness." source
Any performance tests that haven't taken that into account somehow can't be taken too seriously sadly, it's a difficult thing to deal with for review, much like a fresh Vista original release, though at least SP1 shouldn't blank out your index system's index, and cause that to re-catalog everything too, that really would cripple immediate post-install tests.
Your mileage may vary. I have a P4 2Ghz 1GB RAM machine that runs Vista as fast as it's 2Ghz centrino duo 1GB RAM cousin. Only difference: With Vista I get more eye candy and a shorter startup time.
:)
No, no fat ladies for me
http://blogs.technet.com/markrussinovich/archive/2008/02/04/2826167.aspx
Whoa, a lot of people dont get this at all. Here's a hint.
My wife and I recently took over a business that was still using Windows 2000. I had forgotten that a Windows computer could handle email, web-browsing, and QuickBooks just fine on a 6 year old computer with 512MB of RAM.
As Mark said, there were several problems with the XP model. The biggest problem being that large file copy operations could use up all the memory in the system. There were also scenarios where there as double-caching going on.
In Vista RTM, they completely did away with most cached i/o and increased the read/write sizes. This resulted in both a real and a perceived performance penalty for some local copy scenarios, but it dramatically improved network throughput and utilization.
In Vista SP1, they went back to doing *some* cached i/o in certain scenarios. So it's basically a blended approach. They also eliminated the double caching that sometimes took place.
If I read the article correctly, it takes 348 seconds to transfer 1.9GB of data. That amounts to 5.6 MB/sec copyspeed, or about 11.2 MB/s transfer speed on the disk (read + write). A simple, $50 SATA-II disk is able to sustain 50MB/s transfers, read or write, and quality hard disks even more. What is happening with the remaining bandwidth? There is some seek overhead, directory updates, etc but nothing that would slow it down. Also, 11MB/s is hardly a big strain for main memory, cache or PCI bus bandwidth, so it should not affect responsiveness at all. Somebody mentioned lack of rigorous benchmarking because no variance was measured. In this case, it seems many times too slow compared to the physical limit of the disk, so something is fundamentally wrong, irrespective of variance.
I quickly tested this on a SuSE linux machine, and found copy speeds of about 19 MB/sec including syncing to disk (so not tainted by buffering), or 38.2 MB/sec total disk transfer. Accounting for seek overhead, directory updates, etc, that feels like it is limited by the hardware (about 50MB/s for sequential access on this computer). Vista seems to lose about a factor of 4 relative to the hardware. Given the speed of the machine used (cpu, memory, videocard etc) any gui-aspects should not be the limiting factor. All other factors such as different filesystem etc should likewise have a negligable influence. I guess I'll stick to linux for the moment for my IO-intensive work...