We now have a lot of infrastructure for evaluating and improving boot
time. But all of the evaluation we've done has been under ideal
situations - namely on filesystems that readahead well with a
pre-populated ureadahead pack.
But when people try Ubuntu for the first time, their experience won't
be so idealized - they will be running the live session of of the
installer. While it was easy to dismiss this scenario in the past as
unfixable given the slow speed of optical drives, I think we should
look at it more seriously now that USB drives and DVDs are becoming
more popular as an installation mechanism.
As a quick test, I measured the boot time of today's live CD build,
compared with an installed system with a primed ureadahead pack. For
both tests, I used the same desktop-grade hard drive connected over
eSATA, so media speed shouldn't be relevant to the results. The
results are in http://web.mit.edu/broder/Public/livecd-bootchart/ -
you can see that the pre-install live CD boot is about 25% slower.
There are some ongoing initiatives that will make this better - things
like Upstart-in-initramfs will hopefully let us parallelize the
highly serial setup that casper is doing. But at the very least, I
would like for us to start tracking live session boot performance in
the same way we already track post-install boot times, so that we
remain aware of it as an issue.
On Tue, Oct 4, 2011 at 4:51 AM, Sebastien Bacher seb...@ubuntu.com wrote:
Hey,
Since we have been a bit sloppy on that topic since lucid (or busy with
new features) we should aim back at better performances for the lts,
that include cpu and memory usage and login time.
Cheers,
Sebastien Bacher
--
ubuntu-desktop mailing list
ubuntu-desktop@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-desktop
bootchartify
Description: Binary data
--
ubuntu-desktop mailing list
ubuntu-desktop@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-desktop
