First steps and Gnome-startup visualization

Hi all,
My project was accepted for Google Summer of code 2006. Wahoo!

I'll be looking at optimizing the startup performance of Nautilus to both make it faster to start and easier to profile. I've been around the block a bit for profiling (Buy my book, wink, wink... ) , and I've been wanting to do some optimization for GNOME for a while. Now I have a good reason, sweet.

In any event, while I wait for jhbuild to complete on my dodgy internet connection, I have been playing with bootchart to see what the system is doing when gnome starts up.

A lot of work had been done to optimize the system startup time using bootchart, but it can also be used to track the startup of anything. (In this case, I track "startx")

Here's how to do it:
1) Download and build the rpm. (You'll need a bunch of java libraries)

2) Install the RPM.

3) reboot your machine, and start in init level 3. (pass "3" to the boot parameters or modify your inittab)

4) log in as root

(We'll now gather the "cold disk cache" statistics)
5) Tell bootchart to monitor startx and any processes that it generates: "bootchartd start startx"

6) Log out as soon as possible.

7) Run "bootchart". This will generate an SVG file named "./bootchart.svgz" which contains the graph of the startup time.

(Now we'll gather the "warm disk cache" statistics.)

8) Tell bootchart to monitor startx and any processes that it generates: "bootchartd start startx"

9) Log out as soon as possible.

10) Run "bootchart". This will generate an SVG file named "./bootchart.svgz" which contains the graph of the startup time.

Here is an example from FC4:

FC4 cold disk cache:















FC4 warm disk cache:

















Observations:
1) esd seems to start when we try to logout. That's pretty annoying, and I'm sure it is a bug.
2) In the warm cache case, xkbcomp does a lot of disk I/O.

Here's the same thing for FC5:
FC5 gnome startup (cold disk cache):


FC5-Gnome startup (warm disk cache):




Observations:

1) gconfd-2 chews up a chunck of CPU time
2) The esd bug still appears (look at the cold cache chart)

As I see more things, I'll post them.