Ubuntu 11.04 Boot Performance Compared To Past Releases
Phoronix: Ubuntu 11.04 Boot Performance Compared To Past Releases
Now that we have looked at the Ubuntu power consumption going back as far as Ubuntu 8.04 LTS (and found serious power regressions), the next round of testing is providing the Bootchart results for five different systems also going back as far as Ubuntu 8.04 LTS.
These are not regressions. Software regressions are bugs; these are simply unfortunate side-effects from using a new desktop environment, namely Unity.
Interesting info otherwise, though. Thanks.
how can it be disk throughput is not related to boot performance? anyway...
i have filed a bug report months ago...
and there are many things on the net about crappy disk performance on linux
This is why they should have adopted systemd.
And combined with Gnome3, this is why Red Hat with Fedora 15 has utterly out-engineered Canonical once again.
Michael, you should have linked an older article about how they were targeting a 10s boot time.
On a more serious note, when all laptops boot the same, from Atom to Core2, HDD or SSD, something's not right.
In my opinion, users really wanting fast boot times need to customize their system. A general-towarded distribution will always have useless things (for some users). With a custom kernel, only essential services and a fast SSD my system takes 7s from grub to a full kde4 session.
On a side note, if cpu frequency scaling is enabled (as on most laptops), in my experence boot times will typically double, as the cpu boots in the lowest power state.
I benchmarked my fastest system with a Kanotix 64 stock install onto my new ssd without swap and with 6 gb swap (thats what u would create with auto settings). As it has got 6 gb ram it got that much swap. The differences are realy extreme...
and now with 6 gb swap
it seems that the system waits for the activation of swap. After that test i tried to remove unused services got get an even higher bootspeed. Removing of lvm, cryptsetup, dmraid, mdadm, plymouth gave the hugest boost, the rest only minimal. Well huge is a bit overrated for that system, thats the fastest i got when i tried:
I also noticed that bootchart (1) does not go over 90 mb/s, therefore i tried bootchart2 as well but it did not stop at the right time, it was a bit higher the value first and a bit later it did not stop at all, did not debug that case however. But there i saw thruput rates over 200 mb/s (when it ran too long). When i wanted to benchmark natty the bootchart tool from universe did not track anything, it has got no /sbin/bootchartd there, so in theory it should be activated via the initramfs - at least it looked like that. But i never got profiling data... What was used for natty and how was it activated?
Ridiculous, I can boot Linux in 7 second (to the desktop, with postponed services once Qt and KDElibs are done loading) on a 5400rpm hdd on my laptop and they can't do it in less then 20 sec on a 3x faster SSD...
All they do in Ubuntu is not needed to boot corectly, udev is nice, but it is possible to replicate its features with a bash script sniffing dmesd hardware initalization until it is ready to start, some for urandom and some other. Using python at boot is also not the best choice in the world, it's nice, but slow to start. GDM is totally useless 95% of the time and is about 30mb. It take 17 lines of C to replicate the core functionalities (user switching and envvar init) or we can use "login", but it need to start quite late in the process. By starting very few services while preloading X is the best. Dbus and sysfs are needed for advanced DE, but not with minimals one.
They like to lose speed with ureadahead. Not sure if this will go away with later boots. Do you see that takes about 10s time?
Now i installed U 11.04 again, ok you don't need to call pybootchartgui manually with that bootchart version, the pngs are already in /var/log/bootchart together with a date + number. But it is about 3-4s slower than Kanotix with swap.