This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
stap_merge benchmarks
- From: Lukas Berk <lberk at redhat dot com>
- To: systemtap <systemtap at sourceware dot org>
- Date: Fri, 6 Aug 2010 10:09:41 -0400 (EDT)
- Subject: stap_merge benchmarks
For those interested,
I tested out various versions the stap_merge.tcl script to see the how long it would take to generate a merged document. Noting that the greatest performance factor was printing to stderr vs an $outfile I came up with these versions.
Script1 - the original script
Script2 - change the -v flag to print to the $outfile (if it exists)
Script3 - Allow the user to decide to print -v to stderr or the $outfile (again, if it exists)
Each script has the seconds it took to complete, and the size of the output file.
input files (MB) | Script 1 (sec/MB) | Script 2 (sec/MB) | Script 3v1 (sec/MB) | Script 3v2 (sec/MB) | Script 1 (sec/MB)
_________________|___________________|___________________|_-v to sterr_________|_-v to $outfile______|_no -v flag_______
116 / 108 | 31 / 221 | 20 / 235 | 44 / 221 | 37 / 235 | 18 / 221
| | | | |
92 / 65 | 30 / 154 | 13 / 164 | 21 / 154 | 28 / 164 | 12 / 154
| | | | |
47 / 28 | 10 / 73 | 7 / 78 | 10 / 73 | 7 / 78 | 5 / 73
| | | | |
100 / 94 | 26 / 190 | 18 / 203 | 31 / 190 | 34 / 203 | 29 / 190
| | | | |
9.2 / 16 | 5 / 24 | 2 / 27 | 4 / 24 | 3 / 27 | 3 / 24
| | | | |
266 / 282 | 58 / 541 | 59 / 567 | 82 / 541 | 67 / 567 | 61 / 541
| | | | |
87 / 100 | 25 / 183 | 20 / 195 | 27 / 183 | 30 / 195 | 28 / 183
| | | | |
80 / 90 | 23 / 167 | 14 / 178 | 25 / 167 | 16 / 178 | 20 / 167
| | | | |
150 / 154 | 75 / 300 | 31 / 317 | 40 / 300 | 35 / 317 | 31 / 300
| | | | |
It seems that Script 2 generally outperforms the other scripts, even the original script with no verbose flag specified, however the output file is larger. Any thoughts? Are these acceptable results?
Lukas Berk