PS3 Performance increase tests

Saw this article on slashdot, and while reading through the full thread this comment stood out from Ingo:

   "i've got no real complaint about ext3 - with the obligatory
    qualification that "noatime,nodiratime" in /etc/fstab is a
    must ...  So for most file workloads we give Windows a
    20%-30% performance edge, for almost nothing. (for RAM-starved
    kernel builds theperformance difference between atime and
    noatime+nodiratime setups is more on the order of 40%)"

I’d consider the PS3 to have RAM constraints. So I tried out the noatime options in my /etc/fstab with some mixed results (below). First, here is what my default fstab looked like:

/dev/sda2        /           ext3       defaults  1 1
/dev/sda3        swap        swap       defaults  0 0

This I changed to:

/dev/sda2        /           ext3       noatime,nodiratime,defaults  1 1
/dev/sda3        swap        swap       noatime,nodiratime,defaults  0 0

So, first thing I did was the test suggested in the thread in compiling the kernel sources:

make distclean
echo 1 > /proc/sys/vm/drop_caches
echo 2 > /proc/sys/vm/drop_caches
echo 3 > /proc/sys/vm/drop_caches
make allnoconfig
time make

The results, minimal change:

defaults: real= 4m2.331s, user 3m27.480s, sys 0m28.352s
noatime:rea= 4m1.176s, user 3m26.680s, sys 0m27.512s

Given the guy who made the test experienced little to no increase with the same test, and Ingo insisted the improvements were significant, I tried a couple more little (very unscientific) tests.

Opening firefox, then closing window once up (time firefox)

default(noncached)- real= 9.259s, user= 2.624s, sys= 0.296s
default(cached)- real= 4.177s, user= 2.52s, sys= 0.308s

noatime(noncached) real=8.887s, user- 2.604s, sys=0.380s
noatime(cached) real= 3.907s, user = 2.504s, sys=0.336s

It “felt” faster, but the numbers are too close together given human reaction time to close the window to be significant.

Then I tried Open Office writer, and the increase was huge:

default = real=27.381s, user 4.512s, sys= 0.804s
noatime = real=15.100s, user=4.352s, sys=0.732s

When the splash screen popped up with noatime, it went right through the loading phase and popped up a working window fairly quickly.

Conclusions:

I will leave noatime in my /etc/fstab. Even though I couldn’t get any numbers to prove it, the system did seem noticeably faster. I’m thinking the numbers are more an indication that I don’t know how to properly test for these performance increases.

Ah well, I’ll follow up when I can do proper testing.