Hi Auke,
On Mon, 2010-03-29 at 09:59 -0700, Auke Kok wrote:
> > If we can get CONFIG_TASKSTATS=y turned on for MeeGo, I'm happy to
> > package & maintain the bootchart2 package - which is prettier of course.
>
> We're currently planning to ship the new bootchart I originally wrote
> (sorry Michael :)) for Moblin.
Well, reading the code it seems you took a somewhat different - albeit
interesting approach. No doubt we should fallback to the schedstat data
as/when taskstats is not present [1]. Having said that - it seems that
schedstat has a granularity of 1ms - which seems somewhat too chunky to
see lots of bitty activity on a fast CPU - bouncing to/fro from gconfd
eg. Taskstats is in theory ns (though it increases in a suspiciously
chunky fashion ;-). I see lots of apparently idle processes using your
bootchart during boot, but the visualisation is quite nice. I still
think it would be better to have taskstats turned on in your kernel.
> With all the moving around going on here I haven't had the time to
> publish the code on a git repo yet, but we'll do that as soon as a
> meego git repo is running.
My views on this sort of Intel internal re-writing are well known.
There is a good way to do this: work with the existing guys on the
various bootchart projects out there, and try to draw them together.
That is the approach I took with bootchart2. It is also far, far less
wasteful of the scarce resources that we (collectively) have today to
work on tooling. You are of course more than welcome to have bootchart2
commit access if you want to get changes in there; I would love to see
that - I'm not precious about ownership.
There is a good way to do software development in community; and IMNSHO
it doesn't look like this. We don't need yet-another bootchart
project[2] there are currently a handful (now +1).
Luckily I am not a volunteer community member, who has a single
much-loved project they want to contribute to MeeGo, and whom would just
wander off in disgust at this point. Conversely - at least you tried
with your README - though strangely no people are named in it except
you: software doesn't write itself ;-).
In the meantime as/when I get cycles, I will try to merge your work
(properly credited etc.), and see if there are any rendering innovations
bootchart2 can re-use. Then I'll submit it for inclusion in MeeGo & see
where we get.
Regards,
Michael.
[1] - TASKSTATS is a horrible interface anyway - and rather too slow to
poll since it screws up accounting for threads. Any kernel geniuses want
to explain how it can be used for a thread-group ? Of course, more
ideally there would be a /proc/taskstats file with all the ns timing
data required for all tasks (including their threads) that can be
accessed with a single open/read/close: and that is consistent and
correct for that open syscall time.
[2] - and of course, this new bootchart doesn't address the nasty
problems of initrds, and the other general linux distro problems that
are out there: to be fair bootchart2 is only half-way towards my current
pipe-dream solution here: using ptrace to grab data out of the logger
post boot :-) Ah - and a real nasty - the parent/child (ppid)
relationship is pointless in a boot-chart, we want "who forked whom" -
which we can get from a probe (it seems); plenty to do there.
--
[email protected] <><, Pseudo Engineer, itinerant idiot
_______________________________________________
MeeGo-dev mailing list
[email protected]
http://lists.meego.com/listinfo/meego-dev
