Difference between revisions of "Build Environment Effort/Scalability"
From Apache OpenOffice Wiki
B michaelsen (Talk | contribs) m |
B michaelsen (Talk | contribs) m |
||
| Line 2: | Line 2: | ||
__TOC__ | __TOC__ | ||
| − | = Parallelization in the current build process = | + | == Parallelization in the current build process == |
| − | = Measure Configuration = | + | === Measure Configuration === |
Hardware: | Hardware: | ||
| Line 17: | Line 17: | ||
* Source: DEV300_m62 | * Source: DEV300_m62 | ||
| − | == Results == | + | === Results === |
| − | === no parallelization === | + | ==== no parallelization ==== |
A <code>build --all</code> took 11281 seconds (3:08:01 hours). Thus the theoretical ideal build time on this four-way system should be 2821 seconds (0:47:01 hours). | A <code>build --all</code> took 11281 seconds (3:08:01 hours). Thus the theoretical ideal build time on this four-way system should be 2821 seconds (0:47:01 hours). | ||
| − | === only parallelization by build.pl === | + | ==== only parallelization by build.pl ==== |
A <code>build --all -P9</code> took 3429 seconds (0:57:09 hours). Compared to the theoretical shortest build time this an overhead of 21.5%. | A <code>build --all -P9</code> took 3429 seconds (0:57:09 hours). Compared to the theoretical shortest build time this an overhead of 21.5%. | ||
Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes: | Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes: | ||
| Line 28: | Line 28: | ||
And here is the timeline: [[Media:Build_Environment_Effort_Scalability_Timelinep9p1.pdf|P9 P1 Timeline]] | And here is the timeline: [[Media:Build_Environment_Effort_Scalability_Timelinep9p1.pdf|P9 P1 Timeline]] | ||
| − | === parallelization by build.pl and dmake === | + | ==== parallelization by build.pl and dmake ==== |
A <code>build --all -P4 -- -P4</code> took 2820 seconds (0:54:40 hours). Compared to the theoretical shortest build time this is an overhead of 16.3%. | A <code>build --all -P4 -- -P4</code> took 2820 seconds (0:54:40 hours). Compared to the theoretical shortest build time this is an overhead of 16.3%. | ||
Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes: | Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes: | ||
Latest revision as of 14:04, 2 November 2009
Contents
Parallelization in the current build process
Measure Configuration
Hardware:
- System: Sun Ultra 24
- CPU: Intel(R) Core(TM)2 Quad CPU Q9650 @ 3.00GHz
- RAM: 8GB
- FS: ext3 relatime 224GB partition filled 90%
Software:
- OS: Ubuntu Linux 9.04
- Compiler: gcc (Ubuntu 4.3.3-5ubuntu4) 4.3.3
- Configure:
./configure --with-use-shell=bash --with-system-libs --without-system-jars --without-system-icu --without-system-agg --without-system-lpsolve --without-system-mspack --disable-mozilla - Source: DEV300_m62
Results
no parallelization
A build --all took 11281 seconds (3:08:01 hours). Thus the theoretical ideal build time on this four-way system should be 2821 seconds (0:47:01 hours).
only parallelization by build.pl
A build --all -P9 took 3429 seconds (0:57:09 hours). Compared to the theoretical shortest build time this an overhead of 21.5%.
Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes:
And here is the timeline: P9 P1 Timeline
parallelization by build.pl and dmake
A build --all -P4 -- -P4 took 2820 seconds (0:54:40 hours). Compared to the theoretical shortest build time this is an overhead of 16.3%.
Here is a graphic showing how much of the build time is spend with a given number of parallel dmake processes:
And here is the timeline: P4 P4 Timeline
