[lfs-dev] Build speeds

Ken Moffat zarniwhoop at ntlworld.com
Sat Jan 31 17:09:24 PST 2015


On Sat, Jan 31, 2015 at 02:20:55PM -0800, James Powell wrote:
> The only way to speed it up would be to use either a faster read/write rate file system, or upgrade hardware.
> 
> Ken are you using JFS with TRIM on the SSD?
> 
 No, ext4 with TRIM.  I started out without TRIM (,discard when
mounting), and separate fstrim.  Gave it one run like that, and only
timed 'make' [ this is a box where -j5, on 4 CPUs, is no faster than
-j4 ] but did not see any benefit.  Then I added the mount option
and did some more thorough testing with the different I/O
schedulers.  Seems to me that for my usage the kernel buffering is
the dominant factor (so, I would need a much faster CPU to see
benefits, I'm already using DDR3 1600 RAM), followed by the SATA
variant - as I think I said, hdparm -tT shows unbuffered reads are
faster with the SSD, but that does not translate into significant
real benefits.

 For the moment, I'm not upgrading - the box is only 2.5 years old,
and is comparatively reliable for testing R600 radeon (the Aruba in
my A4 is sufficiently uncommon that I've had one kernel bug specific
to Aruba in the past, the A4 is underpowered, and I'm currently
running the modesetting driver to get less CPU usage: might get
round to building tools to try to find what uses the CPU in Xorg.bin
from 1.16.3, might not - on the phenom the usage is acceptable, and
in 1.16.99 it is again called just Xorg instead of Xorg.bin).

ĸen
-- 
Nanny Ogg usually went to bed early. After all, she was an old lady.
Sometimes she went to bed as early as 6 a.m.


More information about the lfs-dev mailing list