Sorry for whoever was expecting any result on Thursday - when I plugged in the EeePC the primary and secondary fuses went off end ended up having a (cold +) candlelight dinner and as I had to leave for the weekend I've been able to restore the electricity only today.
So, today I installed Ubuntu 13.04 and SphinUX on the same EeePC.
Short story:
Results: (almost) absolutely no difference (if you exclude the last lines I posted, which add some "mistery" to the whole story) - the performance improvement I saw in the previous test did not exist in this case.
Next: redo tests on previous platform without installation (for confirmation) + install on previous platform + redo tests
Long story:
Platform + SW:
- In both cases an EeePC with an "Intel(R) Atom(TM) CPU N270 @ 1.60GHz"
- Both fs using ext4 for their rootfs with nothing else (no encryption of home dir, no lvm, etc...).
- Both fs mounted with "noatime".
- Both using Xfce4 desktop.
- No external activity (e.g. updates) going on while the tests were running (checked with "htop").
I left the original settings of the OSs, with the exception of the desktop - Ubuntu was ok but SphinUX with all 3D-thingies was unbearable on the EeePC. I therefore installed in both cases Xfce4 and had therefore almost no CPU used for fancy desktop effects nor anything else.
Both OSs were updated to their latest version.
Details:
Ubuntu:
Code:
uname -a
Linux ubuntupc 3.8.0-23-generic #34-Ubuntu SMP Wed May 29 20:24:54 UTC 2013 i686 i686 i686 GNU/Linux
gcc --version
gcc (Ubuntu/Linaro 4.7.3-1ubuntu1) 4.7.3
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
performance
SphinUX
Code:
uname -a
SphinUX sphinux 3.2.32 #1 SMP LSX-arch 1.0.12-1 i686 GNU/Linux
gcc --version
gcc (Debian 4.7.2-5) 4.7.2
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
ondemand
Tests:
In both cases I:
1) used the program I posted in my first post, but this time hardcoded without needing any keyboard input.
2) zipped a completely random 1GB-file created with "dd if=/dev/urandom of=zipme.random bs=64k count=16384".
Test results:
Ubuntu:
Code:
gcc prime.c
time ./a.out
Enter the number of prime numbers required
The last 30000 prime number is:
The last prime number computed was: 350377
real 2m59.861s
user 2m59.620s
sys 0m0.000s
_________________________________________________________________
time zip zippy.zip zipme.random
adding: zipme.random (deflated 0%)
real 3m0.638s
user 2m37.272s
sys 0m7.204s
SphinUX:
Code:
gcc prime.c
time ./a.out
Enter the number of prime numbers required
The last 30000 prime number is:
The last prime number computed was: 350377
real 3m0.525s
user 3m0.291s
sys 0m0.020s
_________________________________________________________________
time zip zippy.zip zipme.random
adding: zipme.random (deflated 0%)
real 3m3.758s
user 2m43.326s
sys 0m8.245s
So, you see that the results practically do not differ.
Ok Ubuntu might have been a tiny bit faster (perhaps because of the fact that it's using the "performance" governor?) but the key-question was if SphinUX was a lot faster than Ubuntu, which resulted this time in an absolute "no".
This would be the end of the story if I didn't see the huge performance improvement on the other PC => what did that mean?
As this benchmark showed no differences, if there is any then it must be because of something which is different between the two platforms.
The first thing that I have to do is therefore to re-validate the old test, this time as antagonist without using Gentoo/Sabayon keeping the same Ubuntu 13 that I used just now => if that one shows again some differences then it might really be that SphinUX has something "different" from the Linux that we all use, on that particular platform.
To end this post, here is something I initially didn't look at:
Code:
nonroot@hostnotebook ~ $ scp zipme.random root@ubuntupc:/root/
root@10.0.123.1's password:
zipme.random 100% 1024MB 2.6MB/s 3.0MB/s 06:34
nonroot@hostnotebook ~ $ scp zipme.random root@sphinux:/root/
root@10.0.123.2's password:
zipme.random 100% 1024MB 2.8MB/s 3.3MB/s 06:08
(the names of the target hosts/PCs were overwritten manually so they might not match anything you see above - don't want to review what I have masked)
You see that the transfer of the 1GB-file over wifi to the EeePC (source=wifi, target=ethernet) when running SphinUX constantly (because I repeated it a couple of time as I could not believe it) takes 30 seconds less then when transferring it to Ubuntu - astonishing, as I expected the wifi (a normal g-wifi) to be the overall "break" - I really did not expect this improvement but as on the EeePC-side such a transfer uses 50% of a single hyperthreaded-CPU if SphinUX acutally did any optimizations in the kernel then it might explain the difference.
Mistery...
Is this like one of the hints that you see at the end of a movie that shows that a second chapter might soon be coming?
In any case: about to install SphinUX + Ubuntu 13 on an external HDD on my "big" PC (the one I use for Steam
)
Cheers