Joerg Sonnenberger <joerg%britannica.bec.de@localhost> writes:
> Doesn't a modern x86 machine benefit stronger from USE_PIGZGZIP=yes than
> from gzip -4?
I really only meant semi-modern, with 2 cores. The following, more or
less:
$ cat comp.tgz > /dev/null; for i in 1 2 3 4 5 6 7 8 9; do \
echo -n "level $i\t\t"; \
zcat comp.tgz | time sh -c "gzip -$i > comp.level$i.tgz"; \
done
with the netbsd-5 comp.tgz (since it was large) led to:
level 1 19.67 real 16.68 user 0.29 sys
level 2 21.34 real 18.16 user 0.34 sys
level 3 24.69 real 22.03 user 0.35 sys
level 4 27.30 real 24.76 user 0.43 sys
level 5 35.93 real 32.28 user 0.46 sys
level 6 52.61 real 49.35 user 0.36 sys
level 7 68.28 real 64.46 user 0.43 sys
level 8 148.60 real 144.21 user 0.29 sys
level 9 340.09 real 331.89 user 0.36 sys
-rw-r--r-- 1 gdt users 272254860 Mar 18 15:09 comp.level1.tgz
-rw-r--r-- 1 gdt users 265125602 Mar 18 15:09 comp.level2.tgz
-rw-r--r-- 1 gdt users 260917250 Mar 18 15:09 comp.level3.tgz
-rw-r--r-- 1 gdt users 250486924 Mar 18 15:10 comp.level4.tgz
-rw-r--r-- 1 gdt users 244010440 Mar 18 15:10 comp.level5.tgz
-rw-r--r-- 1 gdt users 240770889 Mar 18 15:11 comp.level6.tgz
-rw-r--r-- 1 gdt users 239724088 Mar 18 15:13 comp.level7.tgz
-rw-r--r-- 1 gdt users 238854524 Mar 18 15:15 comp.level8.tgz
-rw-r--r-- 1 gdt users 238383900 Mar 18 15:21 comp.level9.tgz
So with perfect parallel speedup one needs 12.5 cores to win with pigz.
One could argue from the above that -9 never makes sense compared to -8,
except for official builds for downloading; 500K cost over 190 seconds.
The above was done on a machine running 5.1ish as amd64, 8G RAM and:
cpu0 at mainbus0 apid 0: Intel 686-class, 3166MHz, id 0x1067a
cpu1 at mainbus0 apid 1: Intel 686-class, 3166MHz, id 0x1067a
which to me is more or less modern.
It seems like pigz is only used in some places, rather than replacing
all invocations of gzip. (That is probably a minor oversight and
irrelevangt to my query.)
Attachment:
pgpG_wdx6Eyif.pgp
Description: PGP signature