Overhead in 14.04 Comparing Xubuntu and Lubuntu

[Aere Greenway]

All:

In testing UbuntuStudio 14.04 (which uses the XFCE desktop), I have been 
observing a surprising amount of overhead on slow machines. This was 
surprising, given how well Lubuntu 14.04 performed doing the same tests.

Since UbuntuStudio includes a lot of KDE libraries, I decided to install 
Xubuntu 14.04 on my 450 megahertz Pentium III (512 MB RAM) machine (an 
HP Vectra).  It took a very long time to install it.

The results of my testing of Xubuntu 14.04 corresponded to the testing I 
did with UbuntuStudio.  It appears to be associated with the XFCE 
desktop environment, and the overhead appears to be graphics-related 
(the Xorg process using most of the CPU).

I don't know why there is such a radical difference between Lubuntu 
(LXDE desktop), and Xubuntu (XFCE desktop), but I am sure of what I 
observed on that particular machine.  These results correspond with 
similar testing I have done on an HP deskpro 933 mhz (512 MB RAM) 
machine, and also on a 1.7 gigahertz (1GB RAM) machine.  But I looked 
most closely on the 450 Mhz (HP Vectra) machine.

Lubuntu Results:

After booting, with the resource monitor applet showing CPU and RAM, 
fairly quickly the CPU usage went down to low values (6%), and stayed 
there.

On starting a terminal window, and executing a "top" command, the CPU 
usage returned to 6%.  I found that the terminal window could be dragged 
around the screen much more easily than I remembered with dragging other 
windows.

I started the Task Manager application, and observed the CPU usage again 
go back down to around 6%.  However, when I tried to drag the Task 
Manager window around the screen, it moved very slowly (consistent with 
my past experience), and the CPU usage was near 100% while moving it.  
But when I released the window (and it stopped moving), CPU usage went 
back to around 6%.

This closer look corresponds to my earlier testing, in which I concluded 
that Lubuntu 14.04 works well on my 450 megahertz, 512 megabyte RAM 
machine, doing MIDI music with software synthesizers (Qsynth and 
Qjackctl, in particular).  It passed my stress-test with a complicated 
MIDI sequence file with no under-runs, which I was particularly pleased 
with.

The only case I encountered overhead, was when I tried to drag windows 
around the screen, and that was so bad that I stopped trying to drag 
windows at all.

Xubuntu Results:

After booting, and signing-in, it takes a long time for the icons to 
appear on the screen.  The CPU-graph applet on the task-bar stayed at 
100%.  When the icons finally appeared on the screen, the CPU-graph 
applet finally went down to 83% usage, with no windows active.

I started up a terminal session, and ran a "top" command in it. That 
showed that the Xorg process was using around 64% of the CPU (it varied 
between 63% and 66%, but most often near 64%).  There was some percent 
going to the CPU-graph (and System Monitor applets) in the task-bar, but 
not anywhere near as much as the Xorg process.

I could drag the terminal session window around the screen with similar 
performance as with Lubuntu).

I fired-up a Task Manager window, but could not identify in it anything 
similar to the Xorg process.  I could drag the Task Manager window 
around the screen, and it seemed to move at the same rate as the 
terminal window (unlike my experience with Lubuntu, where a 
graphics-oriented window moved much more slowly).

With the Task Manager window active, the "top" command running in the 
terminal window showed the Xorg process using around 50% of the CPU (it 
sometimes went higher), but it appeared to be because the Task Manager 
process was using a significant amount of the CPU, and Xorg could no 
longer use as much as it did on an idle system (the total can't exceed 
100%).

My overall impressions of testing both Xubuntu and UbuntuStudio on the 
450 megahertz machine is that it is way too slow to be practical to use 
(as might have been apparent if anyone were listening to the words I was 
muttering while doing the testing).

It appears that the excessive overhead is graphics-related.

Interestingly, at one point in testing, things were going very slow, and 
I needed to do something to break it loose, but I really didn't want to 
terminate Firefox.  I tried 'rolling-up' the Firefox window (leaving 
only its title-bar, which incidentally, was now easy to drag around the 
screen), and things seemed to then move forward in the testing.

In my earlier testing on a 933 megahertz machine using UbuntuStudio, I 
had reported (to the ZynAddSubFX developers) that ZynAddSubFX was no 
longer usable on my 933 megahertz machine.  They did some checking, and 
their testing didn't show any significant increase in the overhead of 
their application.

 From current results, I am now convinced that overhead is 
graphics-related.

Conclusions:

Given the results in testing Xubuntu and UbuntuStudio (both XFCE 
desktop), I can see why Ali insisted that 14.04 will not perform well on 
low-spec machines.  From what Ali has said, he has been testing Xubuntu 
and Ubuntu-Gnome.  I don't know what testing he did with Lubuntu 14.04, 
and I have not yet tested Ubuntu-Gnome on a slower machine.

But my testing of Lubuntu 14.04 on the same low-spec machines (at least, 
the machines I have), shows that the system performs well (until you try 
to drag a window around the screen).

I don't know why there is such a radical difference in the performance 
of the XFCE desktop, and the LXDE desktop.

The Ubuntu Unity desktop performs poorly on my 1.7 gigahertz, 1 gigabyte 
RAM machine, though I am able to easily drag windows around the screen 
(it has a NVIDIA graphics card).  My impressions are (I haven't 
attempted to measure it yet) that it performs more poorly than Windows 7 
on that same machine.

UbuntuStudio on that 1.7 Ghz machine also performs poorly (it worked 
fine in level  13.10), now (with 14.04) it is barely usable.

I think it would be useful for other people with low-spec machines to 
see if they can duplicate my test results.

All of this poor performance is not apparent in fast machines.  It only 
becomes apparent when you test on slower machines.

-- 
Sincerely,
Aere