[systemd-devel] systemd-udevd: excessive I/O usage

Peeters Simon peeters.simon at gmail.com
Mon Jun 4 13:48:04 PDT 2012


2012/6/4 Kok, Auke-jan H <auke-jan.h.kok at intel.com>:
> On Mon, Jun 4, 2012 at 1:14 PM, Peeters Simon <peeters.simon at gmail.com> wrote:
>> 2012/6/4 Kok, Auke-jan H <auke-jan.h.kok at intel.com>:
>>> On Mon, Jun 4, 2012 at 12:49 PM, Peeters Simon <peeters.simon at gmail.com> wrote:
>>>> Hello all.
>>>>
>>>> while trying to optimize my bootup i noted that i could not squeeze it
>>>> under 17seconds userspace
>>>>
>>>> Looking at the bootchart revealed that this is mainly due to
>>>> systemd-udevd consuming 60% of my bootup I/O.
>>>> This means i lose 12 seconds of bootup just to one process.
>>>
>>> can you post the bootchart SVG file?
>>>
>>> I'm assuming you're using this version of bootchart:
>>>
>>> http://foo-projects.org/~sofar/bootchart/bootchart-1.18.tar.gz
>> apparently not, i did just yum install bootchart and it reports to be
>> version 0.14, will take a look at this version.
>>>
>>>> i was wondering whether there is any way to get this more optimized?
>>>> fewer udev rules? more readahead? something else? (i have no idea what
>>>> this I/O is, except for the fact that it is scattered all over the
>>>> place)
>>>>
>>>> this was with systemd 184(from fedora18) on fedora 17.
>>>
>>> The bootchart SVG will tell us most of the stuff that want to know -
>>> so please share it.
>> here you are (this is with fedoras default bootchart, will try the
>> other one later.)
>
> the old version obscures a lot of information - in this case all the IO
> gets attributed to udev while in fact it's probably one of your kernel
> threads. The version of bootchart I maintain should provide a much
> clearer picture of where the bottlenecks are.

well, here it is, it does not really show the I/O activity the old one
shows (it seems like this one only shows throughput)

between 9 and 15 seconds almost nothing seems to happen, almost no
cpu, almost no I/O throughput, only the sound of my harddisk seeking
the whole time, could it be because udev spawns so many children that
want access to the disk simultaneously?

simon


More information about the systemd-devel mailing list