1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
|
# this is a quick and dirty script to run a netperf TCP_RR and
# TCP_STREAM test concurrently to allow one to see the effect of
# buffer bloat on latency. We assume that netperf has been compiled
# with demo-mode enabled via ./configure --enable-demo
NETPERF=`which netperf`
if [ $? -ne 0 ]
then
echo "Unable to find a netperf binary."
exit -1
fi
CHUNK=60
# first, start the TCP_RR test
RR_START=`date +%s`
echo "Starting netperf TCP_RR at $RR_START" | tee bloat.log
# a negative value for the demo interval (-D) will cause netperf to
# make gettimeofday() calls after every transaction. this will result
# in more accurate demo intervals once the STREAM test kicks-in, but a
# somewhat lower transaction rate. not unlike enabling histogram
# mode.
netperf -H $1 -l 7200 -t TCP_RR -D -0.5 -v 2 -- -r 1 2>&1 > netperf_rr.out &
# sleep CHUNK seconds
sleep $CHUNK
# now run the TCP_STREAM test
STREAM_START=`date +%s`
echo "Starting netperf TCP_STREAM test at $STREAM_START" | tee -a bloat.log
netperf -H $1 -l `expr $CHUNK \* 2` -t TCP_STREAM -D 0.25 -v 2 -- -m 1K 2>&1 > netperf_stream.out
STREAM_STOP=`date +%s`
echo "Netperf TCP_STREAM test stopped at $STREAM_STOP" | tee -a bloat.log
# sleep another CHUNK seconds
sleep $CHUNK
pkill -ALRM netperf
RR_STOP=`date +%s`
echo "Netperf TCP_RR test stopped at $RR_STOP" | tee -a bloat.log
RRDTOOL=`which rrdtool`
if [ $? -ne 0 ]
then
echo "Unable to find rrdtool. You will have to post-process the results by hand"
exit 0
fi
MIN_TIMESTAMP=`grep Interim netperf_rr.out | head -1 | awk '{print int($10)}'`
MAX_TIMESTAMP=`grep Interim netperf_rr.out | tail -1 | awk '{print int($10)}'`
MAX_INTERVAL=`grep Interim netperf_rr.out | awk 'BEGIN{max=0.0} ($6 > max) {max = $6}END{print int(max) + 1}'`
LENGTH=`expr $MAX_TIMESTAMP - $MIN_TIMESTAMP`
$RRDTOOL create netperf_rr.rrd --step 1 --start $MIN_TIMESTAMP \
DS:tps:GAUGE:$MAX_INTERVAL:U:U RRA:AVERAGE:0.5:1:$LENGTH
# now fill it
awk -v rrdtool=$RRDTOOL '($1 == "Interim"){printf("%s update netperf_rr.rrd %.3f:%f\n",rrdtool,$10,$3)}' netperf_rr.out | sh
# now post process the tcp_stream test. we could use STREAM_START and
# STREAM_STOP but these will be just a bit more accurate
STREAM_MIN_TIMESTAMP=`grep Interim netperf_stream.out | head -1 | awk '{print int($10)}'`
STREAM_MAX_TIMESTAMP=`grep Interim netperf_stream.out | tail -1 | awk '{print int($10)}'`
STREAM_MAX_INTERVAL=`grep Interim netperf_stream.out | awk 'BEGIN{max=0.0} ($6 > max) {max = $6}END{print int(max) + 1}'`
STREAM_LENGTH=`expr $STREAM_MAX_TIMESTAMP - $STREAM_MIN_TIMESTAMP`
$RRDTOOL create netperf_stream.rrd --step 1 --start $STREAM_MIN_TIMESTAMP \
DS:mbps:GAUGE:$STREAM_MAX_INTERVAL:U:U RRA:AVERAGE:0.5:1:$STREAM_LENGTH
# now fill it
awk -v rrdtool=$RRDTOOL '($1 == "Interim"){printf("%s update netperf_stream.rrd %.3f:%f\n",rrdtool,$10,$3)}' netperf_stream.out | sh
# now graph it. we want to make sure the chart is at least 800 pixels
# wide, and has enough pixels for every data point
WIDTH=$LENGTH
if [ $WIDTH -lt 800 ]
then
WIDTH=800
fi
SIZE="-w $WIDTH -h 400"
# we want to find the scaling factor for the throughput, with the goal
# being that latency can go to the top of the charts and throughput
# will go half-way up
MAXLATMAXBPS=`$RRDTOOL graph /dev/null \
--start $MIN_TIMESTAMP --end $MAX_TIMESTAMP \
DEF:trans=netperf_rr.rrd:tps:AVERAGE \
CDEF:latency=1.0,trans,/ \
VDEF:maxlatency=latency,MAXIMUM \
DEF:mbps=netperf_stream.rrd:mbps:AVERAGE \
CDEF:bps=mbps,2000000,\* \
VDEF:maxbps=bps,MAXIMUM \
PRINT:maxlatency:"%.20lf" \
PRINT:maxbps:"%.20lf" | sed 1d`
# should I check the completion status of the previous command?
# probably :)
SCALE=`echo $MAXLATMAXBPS | awk '{print $2/$1}'`
$RRDTOOL graph bloat.svg --imgformat SVG \
$SIZE \
--lower-limit 0 \
--start $MIN_TIMESTAMP --end $MAX_TIMESTAMP \
-t "Effect of bulk transfer on latency to $1" \
-v "Seconds" \
--right-axis $SCALE:0 \
--right-axis-label "Bits per Second" \
DEF:trans=netperf_rr.rrd:tps:AVERAGE \
CDEF:latency=1.0,trans,/ \
LINE2:latency#00FF0080:"TCP_RR Latency" \
DEF:mbps=netperf_stream.rrd:mbps:AVERAGE \
CDEF:bps=mbps,1000000,\* \
CDEF:sbps=bps,$SCALE,/ \
LINE2:sbps#0000FFF0:"TCP_STREAM Throughput" \
VRULE:${STREAM_START}#FF000080:"TCP_STREAM start" \
VRULE:${STREAM_STOP}#00000080:"TCP_STREAM stop" \
--x-grid SECOND:10:SECOND:60:SECOND:60:0:%X
|
