Re: modprobe slow linear search of /lib/modules/`uname -r`/modules.deps

[William Cohen <wcohen at redhat dot com>]

Tim Bird wrote:
> William Cohen wrote:
> > One of the thing that became
> > apparent was the amount of data read from /lib/modules/`uname
> > -r`/modules.dep and the time reading the data. modprobe is not smart
> > at all about reading the file; it just does a linear search and stops
> > when it finds the first matching entry.
> >
> > Wouldn't it make sense to place the modules that are actually loaded
> > at the beginning of the modules.dep to reduce the amount of data read
> > by modprobe?
> >
> > I wrote a simple script to examine the modules loaded and put the
> > entries in modules.dep at the beginning of the file.  I then replaced
> > the modules.dep with this modified version. I compared the data
> > collected with bootchart.  The boot is a couple seconds faster for the
> > modified version (1:05 vs 1:07) on AMD64 rawhide machine. 
>
> This is interesting.  I'm not sure if there are embedded
> developers using modprobe (you never know), but it still seems
> interesting (both the immediate results and the technique)
> for embedded folk.  Do you mind if I forward this to the
> celinux-dev mailing list?
>  -- Tim
>
> =============================
> Tim Bird
> Architecture Group Chair, CE Linux Forum
> Senior Staff Engineer, Sony Electronics
> =============================

Go ahead and forward the information to the celinux-dev mailing list. I am not 
sure if the particular data would be applicable to it. I hope that embedded 
systems don't have 270K worth of modules.dep. :)

I posted some mail "Boot-time probing with SystemTap" describing some of the 
experiences with bootchart and what things might do to simplify the use of 
systemtap in similar areas.

-Will