I first noticed the X slowdown when I installed F8. I've used every version of Fedora, and I noticed that the user experience with F8 was not as smooth as earlier versions. I spend most of my day hacking code in vi, so 2D video performance is important to me. I care nothing about 3D video.
When I updated my work box (P III - 512 MB) from FC6 to F10, the slowdown was so bad that I was tempted to reinstall FC6. Basically, F10 converted a very usable older box into something that was almost unusable.
So, I decided to collect a few numbers and post them to see if we can get a little discussion going and maybe provide some feedback to some developers who could actually improve things.
I've got boxes at home running FC6, F8, and F10. The FC6 box is an old AMD Athlon 2Ghz 1.5GB memory, the F8 box is a P4 3 Ghz 2GB memory, and the F10 box is an Intel Core2 2.2 Ghz 4GB memory.
For the test, I created a 1,000,000 line (80 byte lines, 80 MB) text file and timed "cat file" on all the boxes, with and without X. I ran the test several times and reported the fastest time. I also tried turning off anti-aliased text, but that was actually slower.
Bottom line, FC6 running X was 6 times faster than F8 and over 8 times faster than F10. I know that there have been many, many improvements in Fedora over the years, but X looks like it's taking a big step backward. And don't tell me the eye-candy is much better, because I don't care.
If I'm full of shit, please let me know how I can get better X performance.
Maybe someone with ATI video could run the test and post some results.
Regards,
John
FC6 - KDE 3.5.7-9 FC6 - Nvidia video "nv" driver - LCD 1280 x 1024
Console - No X
real 0m31.252s user 0m0.000s sys 0m31.245s
X - KDE konsole - 24 x 80
real 0m33.218s user 0m0.027s sys 0m0.853s
F8 - XFCE 4.6 - Nvidia video "nv" driver - LCD 1680 x 1050
Console - No X
real 0m27.184s user 0m0.003s sys 0m27.178s
X - xfce console - 24 x 80 - LCD 1920 x 1080
real 3m22.129s user 0m0.023s sys 0m0.830s
F10 - XFCE 4.4 - Intel video
Console - No X
real 1m40.978s user 0m0.004s sys 1m40.956s
X - xfce console - 24 x 80
real 4m53.576s user 0m0.000s sys 0m0.036s
I've got boxes at home running FC6, F8, and F10. The FC6 box is an old AMD Athlon 2Ghz 1.5GB memory, the F8 box is a P4 3 Ghz 2GB memory, and the F10 box is an Intel Core2 2.2 Ghz 4GB memory.
For the test, I created a 1,000,000 line (80 byte lines, 80 MB) text file and timed "cat file" on all the boxes, with and without X. I ran the test several times and reported the fastest time. I also tried turning off anti-aliased text, but that was actually slower.
Bottom line, FC6 running X was 6 times faster than F8 and over 8 times faster than F10. I know that there have been many, many improvements in Fedora over the years, but X looks like it's taking a big step backward. And don't tell me the eye-candy is much better, because I don't care.
You seem to have erroneously posted a mix of numbers mixing up version, kernel, X server and hardware. Unless you hold three of those constant to get the variation in the fourth your data is totally meaningless - even if there is a real slow down.
Alan Cox wrote:
I've got boxes at home running FC6, F8, and F10. The FC6 box is an old AMD Athlon 2Ghz 1.5GB memory, the F8 box is a P4 3 Ghz 2GB memory, and the F10 box is an Intel Core2 2.2 Ghz 4GB memory.
For the test, I created a 1,000,000 line (80 byte lines, 80 MB) text file and timed "cat file" on all the boxes, with and without X. I ran the test several times and reported the fastest time. I also tried turning off anti-aliased text, but that was actually slower.
Bottom line, FC6 running X was 6 times faster than F8 and over 8 times faster than F10. I know that there have been many, many improvements in Fedora over the years, but X looks like it's taking a big step backward. And don't tell me the eye-candy is much better, because I don't care.
You seem to have erroneously posted a mix of numbers mixing up version, kernel, X server and hardware. Unless you hold three of those constant to get the variation in the fourth your data is totally meaningless - even if there is a real slow down.
regardless the difference between X and no X in each version/hardware is astounding! what are the reasons for this, your not telling me that X is chewing up that much cpu cycles to turn a 1minute 40sec operation into a 4minute 53sec one?
i think this merits much further analysis!
regardless the difference between X and no X in each version/hardware is astounding! what are the reasons for this, your not telling me that X is chewing up that much cpu cycles to turn a 1minute 40sec operation into a 4minute 53sec one?
Why do you assume this has anything to do with CPU cycles and not graphics performance ?
i think this merits much further analysis!
Feel free, but you need to do it in a disciplined repeatable one change at a time fashion.
Alan
On Tuesday, Mar 24th 2009 at 08:14 -0000, quoth Alan Cox:
=>> regardless the difference between X and no X in each version/hardware is =>> astounding! what are the reasons for this, your not telling me that X is =>> chewing up that much cpu cycles to turn a 1minute 40sec operation into a =>> 4minute 53sec one? => =>Why do you assume this has anything to do with CPU cycles and not =>graphics performance ? => =>> i think this merits much further analysis! => =>Feel free, but you need to do it in a disciplined repeatable one change =>at a time fashion. => =>Alan
Performance issues are very touchy feely, and Alan is correct. One change at a time is the only proper way to measure any kind of performance metric.
Back in an old AI class a few decades ago, Patrick Henry Winston had a quote in his AI book that stuck with me.
Michaels Theorem: You can't learn anything unless you almost already know it.
Measure something. Change one thing, and measure again.
Alan Cox wrote:
regardless the difference between X and no X in each version/hardware is astounding! what are the reasons for this, your not telling me that X is chewing up that much cpu cycles to turn a 1minute 40sec operation into a 4minute 53sec one?
Why do you assume this has anything to do with CPU cycles and not graphics performance ?
i think this merits much further analysis!
Feel free, but you need to do it in a disciplined repeatable one change at a time fashion.
Alan
You're correct of course. I posted some observations and I have hypothesized that I am seeing an X performance regression. I was hoping that someone would have already done the hard work and would have an explanation for what I'm seeing.
The next step is to measure the performance of different versions of X on an otherwise stable platform. That's what I'll be doing this weekend. Now I just need to figure out how to get "old" X running on F10.
Regards,
John