?????: modprobe woes

[Vladimir Dronnikov]

Thanks for testing, Timo!


> Which is still slightly worse than module-init-tools. And I think I can
> beat
> even module-init-tools. I can get rid of most of the mallocs and
> reading can be speed up too by parsing by tokens instead of lines.
>

It is better to read a file in one go (as was proposed but patch was
blocked). Anyway we have to keep all data, have we read it line-by-line and
xstrdup()ed or have we just set pointers to specific locations of one big
memory chunk.
And we could mmap() modules.dep.bb.
And we could use it in binary form with spaces pre-substituted with NULs.
There ia a bunch of potential speedups.


>
> The reason for module-init-tools being faster is for one that all the
> aliases and symbol aliases are in different files. They are not read at all
> if they are not required. Thus your single file approach is bound to be
> slower.
>
>
IMO, 50 modprobes at bootup are tightly related to these symbol:*
definitions. Not an issue comparing with aforementioned improvements in
speed, IMO.

Let us debug the engine, and then we can create interface between vanilla
modules.* and modules.dep.bb so users won't get confused.

-- 
Vladimir
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.busybox.net/pipermail/busybox/attachments/20080806/1216b814/attachment-0002.htm