diff options
| author | Andras Timar <andras.timar@collabora.com> | 2022-01-22 22:00:56 +0100 |
|---|---|---|
| committer | Andras Timar <andras.timar@collabora.com> | 2022-02-13 09:39:00 +0100 |
| commit | 04ae86ce140985f314f7d2312fa101b55f5ef26c (patch) | |
| tree | 3c8ee8816bfb4b98b02eae2d6a08ee184ea87a21 /solenv | |
| parent | 77a3333b45e10959863e2980a7b17caaeac2694f (diff) | |
change rpm compression algorithm to multi threaded xz
Our baseline is CentOS 7, it's safe to switch to xz that is supported there.
Here is my measurement on rpm package build time and package size on a
48 core machine, however 48 cores were never used (8-10 max). Compared to
reference w9.gzdio which was used before the patch.
Compression Build Package
method time % size %
w9.gzdio 100% 100%
w7T.xzdio 69% 66%
w6T.xzdio 62% 67%
w5T.xzdio 55% 67%
w4T.xzdio 41% 70%
w3T.xzdio 31% 73%
w2T.xzdio 26% 75%
w1T.xzdio 24% 77%
Conclusion: we'd like to optimize rather for the build time, therefore
w1T.xzdio was picked. The package size was still smaller than the reference
size, and the build time was 4x faster (in this measurement 18m16 -> 4m26).
That makes a difference.
Change-Id: Iabb532628d38a2d5091fa41fd08bfee69007d0ca
Diffstat (limited to 'solenv')
| -rw-r--r-- | solenv/bin/modules/installer/epmfile.pm | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/solenv/bin/modules/installer/epmfile.pm b/solenv/bin/modules/installer/epmfile.pm index ecddeb395712..495366823604 100644 --- a/solenv/bin/modules/installer/epmfile.pm +++ b/solenv/bin/modules/installer/epmfile.pm @@ -1193,7 +1193,7 @@ sub set_autoprovreq_in_specfile close (FINDREQUIRES); } - $autoreqprovline .= "%define _binary_filedigest_algorithm 1\n%define _binary_payload w9.gzdio\n"; + $autoreqprovline .= "%define _binary_filedigest_algorithm 1\n%define _binary_payload w1T.xzdio\n"; for ( my $i = 0; $i <= $#{$changefile}; $i++ ) { |