Ubuntu 11.10 vs. Fedora 16: Boot Speed, Power Consumption
Phoronix: Ubuntu 11.10 vs. Fedora 16: Boot Speed, Power Consumption
In this article is the first of several articles comparing the recently released Fedora 16 to Ubuntu 11.10. This first article is looking at the boot performance and power consumption from several different notebooks when performing clean installs of Fedora Verne and Ubuntu Oneiric Ocelot.
It'd be great if you could post larger versions of the bootcharts so we could actually see what's going on in the boot tests. No matter how hard I squint I can't quite read most of those bars.
The numbers you got are interesting; F16 seems to boot a lot faster on both my systems (under 10 secs in each case by bootchart's numbers), but they're both running SSDs.
20W on an EEE PC? I thought those things were meant to be quite efficient. Nouveau driver all round; does that W510 have switchable graphics so you can test with intel graphics too?
So Fedora blows the doors off Ubuntu in overall jaw-dropping performance. Nice.
As far as the boot time wrt systemd benefits; well... one of the key things to consider is that systemd is a massive change that is being done in small and manageable steps. They haven't even begun to focus on the performance aspects yet.
The first step in the transition is to get systemd installed and loading the old service files.
The next step in the transition is to convert the old service files to systemd <-- they are currently in the process of this.
AFTER the first two steps are COMPLETED, then you get to start optimizing for performance.
So, no surprise that Fedora was a tad slow in the bootup.
Fedora 16 tested in the stock configuration
I strongly object to the results of the boot time test, in particular the Fedora 16 slowness.
While I can say nothing about Ubuntu (being a RH user since the very first Fedora release many years ago, yes, back in the time when PPP configuration required a bit of black magic), I can say something about optimising Fedora.
1)They do, for some reason, turn on all the possible services by default, part of which looks (to me) like pure admin stuff. E.g. nfs (1-2 services), sendmail (2 services - sendmail && sm-client), rpc (~3-4 services - rpcbind et al). Bluetooth is also on by default, as is CUPS. Now I think these should be off by default, and if the system finds a bluetooth device or a printer on first boot, then, and only then, the service should loaded. Turning off all these things speeds up the boot process quite a lot.
2)The author of the article has not mentioned GRUB2 that, while not giving F16 an advantage over Ubuntu (as they've been using it for quite some time already), at least brings it on par with (a few releases old) Ubuntu. Bootchart does not give us a measurement of the pre-kernel boot time, and thus fails to demonstrate the speedup of the POST-screen-to-bootloader transition which GRUB2 does give us. It is a significant speedup, I must tell you.
To the guy who said SysV remnants slow things down:
From what I've figured out, none of the basic default services are SysV anymore. Maybe some of the ones I've listed above are, but they should not be taken into consideration in this test anyway, as I've already said. Yes, there are a lot of services that have yet to be converted, but these are mostly in the 3rd party packages that provide advanced functionality, unnecessary for the regular user. (There is a bug in the RedHat bugzilla that tracks the SysV to systemd progress.)
So to conclude, I'd say that you can make Fedora fast (especially nowadays, with systemd and total SysV->SD conversion), but you have to at least try. Comparing things like this without trying to understand how they work is not the best practice.
Actually it's probably more about comparing Unity and Gnome Shell.
Originally Posted by pszilard
It has some meaning for beginners to compare distros for out of the box performance. But for specialists and developers, it would be welcomed to compare things on comparable basis with least factors changing.
Like in this very informative test :
These results should lead to reproduce these benchmarks against new versions of the desktop managers.
pszilard: honestly, it's long been my belief that there's unlikely to be any real noticeable difference in performance between two distros with similar configurations (same desktop, same services loaded). there just isn't a whole lot of stuff we can do to make our distro magically perform better than anyone else's. we're all building the same code in the end.
performance improvements are much more likely to happen in the kernel modules or mesa anyway, these days.