Configure Cache Effects on Bash Build: Difference between revisions
No edit summary |
No edit summary |
||
(3 intermediate revisions by the same user not shown) | |||
Line 1: | Line 1: | ||
== Building Bash - The experiment == | == Building Bash - The experiment == | ||
Why bash? Because it has enough pieces to make it interesting (~20-30 minutes on my 36 core workstation) and does not build an image, so I can ignore the rpm install/pseudo part of the build. | Why bash? Because it has enough pieces to make it interesting (~20-30 minutes on my 36 core workstation) and does not build an image, so I can ignore the rpm install/pseudo part of the build. | ||
== What I did == | == What I did == | ||
Each build: | |||
<code> | |||
. ./poky/oe-init-build-env | |||
bitbake bash | |||
</code> | |||
The upstream code was in the downloads directory and I <code>rm -rf sstate-cache tmp cache </code> between each build. | |||
<br> | |||
I have 4 builds, 2 with configure cache's for all the recipes and 2 without. I then used <code> buildstats-diff --multi </code> to compare the averages for cache vs non cache. | I have 4 builds, 2 with configure cache's for all the recipes and 2 without. I then used <code> buildstats-diff --multi </code> to compare the averages for cache vs non cache. | ||
To get the recipe specific caches I have the following in a conf/auto.conf file: | To get the recipe specific caches I have the following in a conf/auto.conf file: | ||
Line 66: | Line 73: | ||
</code> | </code> | ||
This isn't how I'd recommend doing this by any means, but a couple of awk/sed scripts let me | This isn't how I'd recommend doing this by any means, but a couple of awk/sed scripts let me extract and automate the creation of the above so it was pretty painless. | ||
== The Results == | == The Results == | ||
=== walltime === | === walltime === | ||
<code> | <code> | ||
$ buildstats-diff --multi --diff-attr walltime tmp.bash.nocache.2runs/buildstats tmp.bash.cache.2runs/buildstats | |||
Ignoring tasks less than 00:05.0 (5.0s) | Ignoring tasks less than 00:05.0 (5.0s) | ||
Ignoring differences less than 00:02.0 (2.0s) | Ignoring differences less than 00:02.0 (2.0s) | ||
Line 149: | Line 157: | ||
=== cputime === | === cputime === | ||
<code> | <code> | ||
$ buildstats-diff --multi --diff-attr cputime tmp.bash.nocache.2runs/buildstats tmp.bash.cache.2runs/buildstats | |||
Ignoring tasks less than 00:03.0 (3.0s) | Ignoring tasks less than 00:03.0 (3.0s) | ||
Ignoring differences less than 00:01.0 (1.0s) | Ignoring differences less than 00:01.0 (1.0s) |
Latest revision as of 06:18, 10 November 2016
Building Bash - The experiment
Why bash? Because it has enough pieces to make it interesting (~20-30 minutes on my 36 core workstation) and does not build an image, so I can ignore the rpm install/pseudo part of the build.
What I did
Each build:
. ./poky/oe-init-build-env
bitbake bash
The upstream code was in the downloads directory and I rm -rf sstate-cache tmp cache
between each build.
I have 4 builds, 2 with configure cache's for all the recipes and 2 without. I then used buildstats-diff --multi
to compare the averages for cache vs non cache.
To get the recipe specific caches I have the following in a conf/auto.conf file:
EXTRA_OECONF_append_pn-gettext :=" --cache-file=/YP/config-cache/gettext-0.19.8.1.cache"
EXTRA_OECONF_append_pn-libffi :=" --cache-file=/YP/config-cache/libffi-3.2.1.cache"
EXTRA_OECONF_append_pn-libtool-cross :=" --cache-file=/YP/config-cache/libtool-cross-2.4.6.cache"
EXTRA_OECONF_append_pn-gdbm :=" --cache-file=/YP/config-cache/gdbm-1.12.cache"
EXTRA_OECONF_append_pn-flex :=" --cache-file=/YP/config-cache/flex-2.6.0.cache"
EXTRA_OECONF_append_pn-sed :=" --cache-file=/YP/config-cache/sed-4.2.2.cache"
EXTRA_OECONF_append_pn-sqlite3 :=" --cache-file=/YP/config-cache/sqlite3-3.14.1.cache"
EXTRA_OECONF_append_pn-ncurses :=" --cache-file=/YP/config-cache/ncurses-6.0+20160625.cache"
EXTRA_OECONF_append_pn-python :=" --cache-file=/YP/config-cache/python-2.7.12.cache"
EXTRA_OECONF_append_pn-bash :=" --cache-file=/YP/config-cache/bash-4.3.30.cache"
EXTRA_OECONF_append_pn-diffutils :=" --cache-file=/YP/config-cache/diffutils-3.4.cache"
EXTRA_OECONF_append_pn-glibc :=" --cache-file=/YP/config-cache/glibc-2.24.cache"
EXTRA_OECONF_append_pn-expat :=" --cache-file=/YP/config-cache/expat-2.2.0.cache"
EXTRA_OECONF_append_pn-make :=" --cache-file=/YP/config-cache/make-4.2.1.cache"
EXTRA_OECONF_append_pn-m4 :=" --cache-file=/YP/config-cache/m4-1.4.17.cache"
EXTRA_OECONF_append_pn-bc :=" --cache-file=/YP/config-cache/bc-1.06.cache"
EXTRA_OECONF_append_pn-bzip2 :=" --cache-file=/YP/config-cache/bzip2-1.0.6.cache"
EXTRA_OECONF_append_pn-readline :=" --cache-file=/YP/config-cache/readline-6.3.cache"
EXTRA_OECONF_append_pn-gawk :=" --cache-file=/YP/config-cache/gawk-4.1.3.cache"
EXTRA_OECONF_append_pn-db :=" --cache-file=/YP/config-cache/db-6.0.35.cache"
EXTRA_OECONF_append_pn-rpm-native :=" --cache-file=/YP/config-cache/rpm-native-5.4.16.cache"
EXTRA_OECONF_append_pn-cross-localedef-native :=" --cache-file=/YP/config-cache/cross-localedef-native-2.24.cache"
EXTRA_OECONF_append_pn-readline-native :=" --cache-file=/YP/config-cache/readline-native-6.3.cache"
EXTRA_OECONF_append_pn-attr-native :=" --cache-file=/YP/config-cache/attr-native-2.4.47.cache"
EXTRA_OECONF_append_pn-libmpc-native :=" --cache-file=/YP/config-cache/libmpc-native-1.0.3.cache"
EXTRA_OECONF_append_pn-db-native :=" --cache-file=/YP/config-cache/db-native-6.0.35.cache"
EXTRA_OECONF_append_pn-m4-native :=" --cache-file=/YP/config-cache/m4-native-1.4.17.cache"
EXTRA_OECONF_append_pn-binutils-native :=" --cache-file=/YP/config-cache/binutils-native-2.27.cache"
EXTRA_OECONF_append_pn-gettext-native :=" --cache-file=/YP/config-cache/gettext-native-0.19.8.1.cache"
EXTRA_OECONF_append_pn-mpfr-native :=" --cache-file=/YP/config-cache/mpfr-native-3.1.4.cache"
EXTRA_OECONF_append_pn-byacc-native :=" --cache-file=/YP/config-cache/byacc-native-20160606.cache"
EXTRA_OECONF_append_pn-gperf-native :=" --cache-file=/YP/config-cache/gperf-native-3.0.4.cache"
EXTRA_OECONF_append_pn-bison-native :=" --cache-file=/YP/config-cache/bison-native-3.0.4.cache"
EXTRA_OECONF_append_pn-prelink-native :=" --cache-file=/YP/config-cache/prelink-native-1.0+gitAUTOINC+ef20628dd7.cache"
EXTRA_OECONF_append_pn-sqlite3-native :=" --cache-file=/YP/config-cache/sqlite3-native-3.14.1.cache"
EXTRA_OECONF_append_pn-python-native :=" --cache-file=/YP/config-cache/python-native-2.7.12.cache"
EXTRA_OECONF_append_pn-libpcre-native :=" --cache-file=/YP/config-cache/libpcre-native-8.39.cache"
EXTRA_OECONF_append_pn-gcc-cross-initial-x86_64 :=" --cache-file=/YP/config-cache/gcc-cross-initial-x86_64-6.2.0.cache"
EXTRA_OECONF_append_pn-flex-native :=" --cache-file=/YP/config-cache/flex-native-2.6.0.cache"
EXTRA_OECONF_append_pn-util-macros-native :=" --cache-file=/YP/config-cache/util-macros-native-1.19.0.cache"
EXTRA_OECONF_append_pn-gmp-native :=" --cache-file=/YP/config-cache/gmp-native-6.1.1.cache"
EXTRA_OECONF_append_pn-elfutils-native :=" --cache-file=/YP/config-cache/elfutils-native-0.166.cache"
EXTRA_OECONF_append_pn-pkgconfig-native :=" --cache-file=/YP/config-cache/pkgconfig-native-0.29.1+gitAUTOINC+87152c05be.cache"
EXTRA_OECONF_append_pn-ossp-uuid-native :=" --cache-file=/YP/config-cache/ossp-uuid-native-1.6.2.cache"
EXTRA_OECONF_append_pn-file-native :=" --cache-file=/YP/config-cache/file-native-5.28.cache"
EXTRA_OECONF_append_pn-bzip2-native :=" --cache-file=/YP/config-cache/bzip2-native-1.0.6.cache"
EXTRA_OECONF_append_pn-binutils-cross-x86_64 :=" --cache-file=/YP/config-cache/binutils-cross-x86_64-2.27.cache"
EXTRA_OECONF_append_pn-quilt-native :=" --cache-file=/YP/config-cache/quilt-native-0.64.cache"
EXTRA_OECONF_append_pn-xproto-native :=" --cache-file=/YP/config-cache/xproto-native-7.0.31.cache"
EXTRA_OECONF_append_pn-acl-native :=" --cache-file=/YP/config-cache/acl-native-2.2.52.cache"
EXTRA_OECONF_append_pn-xz-native :=" --cache-file=/YP/config-cache/xz-native-5.2.2.cache"
EXTRA_OECONF_append_pn-popt-native :=" --cache-file=/YP/config-cache/popt-native-1.16.cache"
EXTRA_OECONF_append_pn-makedepend-native :=" --cache-file=/YP/config-cache/makedepend-native-1.0.5.cache"
EXTRA_OECONF_append_pn-expat-native :=" --cache-file=/YP/config-cache/expat-native-2.2.0.cache"
EXTRA_OECONF_append_pn-automake-native :=" --cache-file=/YP/config-cache/automake-native-1.15.cache"
EXTRA_OECONF_append_pn-autoconf-native :=" --cache-file=/YP/config-cache/autoconf-native-2.69.cache"
EXTRA_OECONF_append_pn-gcc-cross-x86_64 :=" --cache-file=/YP/config-cache/gcc-cross-x86_64-6.2.0.cache"
EXTRA_OECONF_append_pn-libtool-native :=" --cache-file=/YP/config-cache/libtool-native-2.4.6.cache"
EXTRA_OECONF_append_pn-ncurses-native :=" --cache-file=/YP/config-cache/ncurses-native-6.0+20160625.cache
This isn't how I'd recommend doing this by any means, but a couple of awk/sed scripts let me extract and automate the creation of the above so it was pretty painless.
The Results
walltime
$ buildstats-diff --multi --diff-attr walltime tmp.bash.nocache.2runs/buildstats tmp.bash.cache.2runs/buildstats
Ignoring tasks less than 00:05.0 (5.0s)
Ignoring differences less than 00:02.0 (2.0s)
PKG TASK ABSDIFF RELDIFF WALLTIME1 -> WALLTIME2
gettext-native do_configure -63.7s -33.6% 189.8s -> 126.1s
gettext do_configure -45.5s -29.5% 154.0s -> 108.5s
bison-native do_configure -41.2s -61.4% 67.2s -> 26.0s
m4-native do_configure -40.3s -79.7% 50.6s -> 10.3s
rpm-native do_configure -36.2s -34.1% 106.2s -> 70.0s
ncurses-native do_configure -26.4s -70.1% 37.6s -> 11.2s
python-native do_configure -21.5s -49.3% 43.7s -> 22.2s
pkgconfig-native do_configure -17.1s -32.7% 52.2s -> 35.1s
db do_configure -16.7s -78.9% 21.2s -> 4.5s
diffutils do_configure -16.6s -39.0% 42.5s -> 25.9s
mpfr-native do_configure -15.5s -54.3% 28.6s -> 13.1s
sed do_configure -13.5s -49.5% 27.2s -> 13.8s
python do_configure -12.6s -36.0% 35.0s -> 22.4s
db-native do_configure -11.9s -73.4% 16.2s -> 4.3s
sqlite3-native do_configure -11.3s -57.9% 19.5s -> 8.2s
m4 do_configure -10.6s -36.8% 28.8s -> 18.2s
bash do_configure -10.0s -53.2% 18.9s -> 8.8s
xproto-native do_configure -9.7s -58.0% 16.7s -> 7.0s
file-native do_configure -9.1s -39.6% 23.0s -> 13.9s
gmp-native do_configure -8.2s -24.1% 34.1s -> 25.8s
readline-native do_configure -7.1s -55.9% 12.7s -> 5.6s
linux-libc-headers do_install -7.0s -35.1% 20.0s -> 13.0s
libtool-native do_configure -6.3s -24.3% 26.0s -> 19.7s
gawk do_configure -6.2s -24.1% 25.8s -> 19.6s
flex do_configure -5.9s -28.0% 20.9s -> 15.1s
libpcre-native do_configure -5.7s -27.1% 21.1s -> 15.4s
elfutils-native do_configure -5.7s -36.1% 15.7s -> 10.0s
ncurses do_configure -5.6s -29.5% 19.0s -> 13.4s
expat-native do_configure -5.5s -31.1% 17.6s -> 12.1s
gettext-native do_compile -4.8s -12.1% 39.3s -> 34.6s
prelink-native do_configure -4.6s -26.8% 17.0s -> 12.5s
gdbm do_configure -4.1s -24.1% 17.2s -> 13.0s
flex-native do_configure -4.1s -20.5% 19.9s -> 15.9s
xz-native do_configure -3.9s -17.5% 22.6s -> 18.7s
gmp-native do_compile -3.7s -30.4% 12.1s -> 8.4s
cross-localedef-native do_configure -3.5s -37.3% 9.5s -> 6.0s
make do_configure -3.2s -27.4% 11.5s -> 8.4s
ossp-uuid-native do_configure -2.9s -23.0% 12.8s -> 9.9s
perl-native do_configure -2.8s -4.5% 62.7s -> 59.9s
libtool-cross do_configure -2.7s -17.6% 15.1s -> 12.5s
ncurses-native do_populate_sysroot -2.5s -30.5% 8.1s -> 5.6s
readline do_configure -2.4s -32.0% 7.6s -> 5.1s
openssl-native do_install -2.3s -4.7% 50.0s -> 47.6s
bash do_package 2.1s +44.6% 4.7s -> 6.8s
gcc-source-6.2.0 do_patch 2.3s +38.9% 6.0s -> 8.4s
bash do_patch 2.5s +50.7% 4.9s -> 7.3s
openssl do_patch 2.5s +49.0% 5.0s -> 7.5s
glibc-locale do_package_write_rpm 2.5s +0.8% 298.4s -> 301.0s
glibc do_unpack 2.8s +33.7% 8.2s -> 11.0s
binutils-native do_unpack 3.0s +38.4% 7.8s -> 10.8s
ncurses-native do_install 3.3s +8.7% 37.8s -> 41.1s
rpm-native do_patch 3.4s +22.8% 15.0s -> 18.4s
elfutils-native do_patch 4.0s +85.7% 4.6s -> 8.6s
db do_unpack 4.1s +271.1% 1.5s -> 5.7s
libtool-native do_compile 4.2s +58.0% 7.2s -> 11.4s
automake-native do_populate_sysroot 4.4s +695.3% 0.6s -> 5.0s
glibc-initial do_patch 4.5s +91.0% 5.0s -> 9.5s
openssl-native do_patch 4.6s +88.2% 5.2s -> 9.7s
binutils-native do_fetch 5.0s +143.5% 3.5s -> 8.5s
cross-localedef-native do_unpack 5.0s +78.2% 6.4s -> 11.4s
perl-native do_compile 5.2s +16.4% 31.7s -> 36.9s
binutils-cross-x86_64 do_compile 6.7s +11.0% 60.7s -> 67.4s
gcc-source-6.2.0 do_unpack 7.8s +18.6% 42.2s -> 50.0s
binutils-native do_compile 8.3s +14.6% 57.2s -> 65.5s
ncurses-native do_compile 9.1s +80.1% 11.4s -> 20.5s
gettext-native do_install 13.1s +68.6% 19.0s -> 32.1s
Cumulative walltime:
-434.5s -7.9% 1:31:26.8 (5486.8s) -> 1:24:12.2 (5052.2s)
cputime
$ buildstats-diff --multi --diff-attr cputime tmp.bash.nocache.2runs/buildstats tmp.bash.cache.2runs/buildstats
Ignoring tasks less than 00:03.0 (3.0s)
Ignoring differences less than 00:01.0 (1.0s)
PKG TASK ABSDIFF RELDIFF CPUTIME1 -> CPUTIME2
gettext-native do_configure -47.1s -38.2% 123.2s -> 76.1s
gettext do_configure -35.0s -29.2% 119.9s -> 84.9s
rpm-native do_configure -29.8s -36.2% 82.4s -> 52.5s
bison-native do_configure -24.8s -53.4% 46.3s -> 21.6s
m4-native do_configure -17.6s -91.3% 19.2s -> 1.7s
python-native do_configure -16.8s -50.8% 33.2s -> 16.3s
diffutils do_configure -14.9s -44.6% 33.4s -> 18.5s
db do_configure -12.3s -82.5% 14.9s -> 2.6s
pkgconfig-native do_configure -11.1s -33.6% 33.1s -> 22.0s
python do_configure -9.9s -38.3% 26.0s -> 16.0s
ncurses-native do_configure -9.8s -67.3% 14.6s -> 4.8s
sed do_configure -9.7s -51.6% 18.8s -> 9.1s
db-native do_configure -8.3s -77.0% 10.8s -> 2.5s
m4 do_configure -7.0s -35.1% 19.8s -> 12.9s
gmp-native do_configure -6.9s -31.8% 21.6s -> 14.7s
bash do_configure -6.6s -54.3% 12.1s -> 5.5s
ncurses do_configure -5.5s -47.2% 11.7s -> 6.2s
file-native do_configure -5.1s -37.3% 13.6s -> 8.5s
xz-native do_configure -4.3s -25.7% 16.9s -> 12.5s
flex-native do_configure -3.7s -26.2% 14.3s -> 10.6s
gawk do_configure -3.7s -20.9% 17.7s -> 14.0s
readline-native do_configure -3.6s -55.1% 6.5s -> 2.9s
mpfr-native do_configure -3.5s -26.9% 13.1s -> 9.6s
flex do_configure -3.2s -22.5% 14.4s -> 11.1s
libtool-native do_configure -3.2s -27.1% 11.8s -> 8.6s
gdbm do_configure -3.1s -23.9% 12.9s -> 9.8s
cross-localedef-native do_configure -3.0s -50.0% 6.0s -> 3.0s
libpcre-native do_configure -2.8s -22.2% 12.6s -> 9.8s
ossp-uuid-native do_configure -2.8s -31.8% 8.7s -> 5.9s
xproto-native do_configure -2.7s -39.3% 6.9s -> 4.2s
makedepend-native do_configure -2.5s -36.9% 6.7s -> 4.2s
make do_configure -2.4s -29.7% 8.2s -> 5.8s
popt-native do_configure -2.2s -20.1% 11.2s -> 8.9s
perl-native do_configure -2.2s -5.6% 39.9s -> 37.7s
elfutils-native do_configure -2.2s -26.6% 8.2s -> 6.0s
sqlite3-native do_configure -2.1s -28.5% 7.3s -> 5.2s
readline do_configure -1.9s -37.4% 5.2s -> 3.2s
db do_package_write_rpm -1.9s -5.9% 32.8s -> 30.8s
libtool-cross do_configure -1.9s -17.9% 10.7s -> 8.8s
prelink-native do_configure -1.9s -17.4% 10.8s -> 8.9s
libffi do_configure -1.8s -15.1% 12.2s -> 10.3s
expat do_configure -1.7s -16.9% 10.0s -> 8.3s
bc do_configure -1.6s -29.1% 5.6s -> 3.9s
sqlite3 do_configure -1.6s -22.4% 7.2s -> 5.5s
gmp-native do_compile -1.6s -3.3% 48.0s -> 46.5s
expat-native do_configure -1.6s -17.0% 9.2s -> 7.6s
perl-native do_install -1.5s -3.8% 40.6s -> 39.1s
libmpc-native do_configure -1.5s -14.3% 10.6s -> 9.1s
glibc do_install -1.3s -0.9% 141.5s -> 140.2s
attr-native do_configure -1.2s -15.7% 7.9s -> 6.7s
bzip2-native do_configure -1.2s -19.6% 6.1s -> 4.9s
gcc-cross-initial-x86_64 do_compile -1.0s -0.1% 824.5s -> 823.4s
db-native do_compile -1.0s -1.6% 61.9s -> 60.9s
bzip2 do_configure -1.0s -15.8% 6.3s -> 5.3s
perl-native do_populate_sysroot 1.0s +12.2% 8.5s -> 9.5s
perl do_compile 1.4s +0.5% 307.1s -> 308.5s
binutils-native do_compile 1.8s +0.4% 442.8s -> 444.6s
openssl do_install 1.9s +4.0% 46.6s -> 48.5s
gcc-runtime do_compile 1.9s +0.9% 213.0s -> 215.0s
ncurses-native do_compile 2.0s +2.0% 102.7s -> 104.8s
ncurses do_compile 2.2s +1.9% 116.4s -> 118.6s
gcc-cross-x86_64 do_compile 2.2s +0.2% 913.4s -> 915.6s
binutils-cross-x86_64 do_compile 2.8s +0.7% 388.3s -> 391.1s
glibc-locale do_package_write_rpm 3.0s +1.0% 294.4s -> 297.5s
glibc-locale do_package 12.9s +3.0% 427.0s -> 439.9s
Cumulative cputime:
-338.7s -3.7% 2:32:59.4 (9179.4s) -> 2:27:20.7 (8840.7s)
As you can see, it was a 4% gain in cputime or an almost 8% gain in walltime. The configure caches will be nowhere near as effective as sstate cache, but they *should* be far less brittle in the face of minor changes. This approach needs more investigation, but I wanted to get some #'s down to quantify what we might see/expect.