I rewired the infrastructure to use sysbench. So the way things now
work is:
1. Run raw sysbench results
2. Run sysbench simultaneously with pmlogger
3. Run sysbench simultaneously with, each in turn, atop, iostat, mpstat,
sar, vmstat
4. The metric logging interval and length of run can be specified. The
length of run is an approximation for sysbench.
So comparing the time of execution results of 2 and 3 with the raw
results from 1 yields.
# atop 0.847166%
# iostat 2.18685%
# mpstat 2.9467%
# pmlogger 0.157024%
# sar -0.458817%
# vmstat 0.101108%
% gnuplot ,gnuplot # graph the above results
cat ,gnuplot
set output "pmbenchmark.ps"
set terminal postscript portrait
set xtics ("raw" 1, "pmlogger" 2, "sar" 3, "iostat" 4, "mpstat" 5,
"vmstat" 6, "atop" 7)
set xlabel "benchmark"
set ylabel "time (s)"
set style line 1 lt rgb "red" lw 20
plot ",pmgnuplot0" with impulses ls 1
cat ,pmgnuplot0
7 1316.59
4 1334.08
5 1344
2 1307.58
1 1305.53
3 1299.54
6 1306.85
|