LFS size and hardware requirements

[Bruce Dubbs]

Alexander E. Patrakov wrote:
> Hello,
> 
> Some newbies get caught by our advertisement (which might be true for older 
> versions of LFS, but is untested as of LFS-6.3):
> 
>> It is not difficult to build an LFS system of less than 100 megabytes (MB),
>> which is substantially smaller than the majority of existing installations.
>> Does this still sound like a lot of space? A few of us have been working on
>> creating a very small embedded LFS system. We successfully built a system
>> that was specialized to run the Apache web server with approximately 8MB of
>> disk space used. Further stripping could bring this down to 5 MB or less. Try
>> that with a regular distribution! This is only one of the many benefits of
>> designing your own Linux implementation.

The above is still true, but perhaps there should be a modification:

"It is not difficult to modify a standard LFS system to use less than 100 
megabytes (MB)..."

This can be done easily by removing /usr/share, /usr/include and perhaps a few 
other files.  Getting it down to less than 50M takes a little more work, but is 
not that hard.  Of course, the lower you go, the more knowledge it takes.

> ...and attempt to build LFS on their slow 586-class computers with only 16 MB of 
> RAM. This is obviously a waste of time, both for them and for us. Additionally, 
> the mentioned 100 MB system obviously contains significant deviations from the 
> book and thus cannot be counted as LFS.

I can't find anywhere where the book refers to a slow 586-class system.   The 
SBU section already says that "Glibc .. could take up to three days on slower 
systems!"

> P.S. I accept the challenge to "try that with a regular distribution".

I am waiting for your results.  Can you do it without breaking updates via their 
package manager?


> Proposal:
> 
> 1) Remove this advertisement.

Disagree.  It is still valid, but could use some tweaks.

> 2) List hardware requirements (CPU, RAM, hard disk space) on the same page as 
> software host requirements, or immediately before it. These requirements should 
> be set so that the total build time (including all testsuites) is less than 8 
> hours, and that the build process never needs to get into swap (the worst case 
> seems to be Chapter 5 gcc Pass1 when starting from a host that is based on gcc-3.3).

Disagree.  The SBU page already can give a user an idea about how long it takes 
for various combinations of hardware.  A user can judge for himself from there.
Again, the SBU page could use some updates.

> 3) When package management enters the book, include a procedure for building 
> packages for a lower CPU (basically, import config.site from the LiveCD and 
> adjust toolchain and perl configure arguments as done there) and transferring 
> LFS and subsequent packages to a different machine.

Disagree.  This is something for a hint or something similar to the user notes 
on BLFS.  Trying to address every corner case in the book would be distracting 
to the majority of users.

   -- Bruce