Re: [Theodore Y. Ts'o: Re: Question about Facility names]
Thanks for Ted Cc'ing me on this. I'm not on the lsb-spec list (I'm
on enough lists already!).
> ------- Forwarded Message
> From: "Theodore Y. Ts'o" <tytso@MIT.EDU>
> There's actually another way these dependencies can be used, and
> happily, either Richard Gooch had the same idea I had, or he read the
> LSB spec and decided to implement it (don't know which). Anyway, a
> Debian developer recently pointed me at this:
>
> http://www.atnf.csiro.au/~rgooch/linux/boot-scripts/
As it says in the first paragraph of this document, these scripts came
about because Patrick (Jordan) and I have bitched and moaned to each
other for years about the way Unix systems boot. In January/February
this year when I was back in Sydney we spent an evening deciding the
One True Way[tm] to boot systems. I coded it up and in March we talked
to Wichert Akkerman about Debian using this scheme. He pushed for the
provide(8) functionality, which I started implementing on the plane to
ALS (yes, it's sad when a plane trip is the first time for months that
I get a chance to do some solid coding).
I hadn't seen the LSB spec. Still haven't actually :-) I'll get around
to that sometime...
> I haven't tried converting a system to use it, but looking at his design
> and the source code this is really cool stuff.
<blush> Note that my WWW page has a link to a tarball which contains
sample boot scripts (the ones I run). Feel free to try them out.
> It might be interesting for folks to look at it and report what they
> think. What Richard did is basically what I always wanted to do,
> but didn't have time to code. (About the only thing which I think
> he hasn't coded was the ability to run boot scripts in parallel to
> speed up the boot process, but that adds a lot of complexity.)
Actually, parallel booting was available back in March. And it
works. The only major development since then has been provide(8).
> Alternatively (and probably the better solution), if we think some
> people will want to stay with System V init scripts, we can modify
> the proposal to make it more simpleinit friendly, while still
> allowing a mapper to map things to the old System V numbers for
> those people who want to stick with the old System V way of doing
> things.
Wichert suggested that the existing Debian boot scripts can have calls
to need(8) and provide(8) added, with the default installation having
symlinks to /bin/true for /sbin/need and /sbin/provide, and then an
optional package which replaces these with the real thing.
Something I'd like to do to speed up booting (I'm basically
seek-limited now) is to have the kernel record a list of blocks it's
read in, and then dump that list upon a certain ioctl(2) or some such.
The boot scripts could then use this is reorder the list of blocks and
write that into a database. Next boot, either init(8) or an early
script reads that database and does dummy reads of those blocks. This
would eliminate seek time overheads for booting.
This would require a kernel hack of course, and I don't know if it
would fly. But I could probably trim boot boot process down to a few
seconds.
Regards,
Richard....
Permanent: rgooch@atnf.csiro.au
Current: rgooch@ras.ucalgary.ca
Reply to: