17
18Please submit patches against the master copy to the tcpdump project on
19sourceforge.net.
20
21formerly from Lawrence Berkeley National Laboratory
22 Network Research Group <tcpdump@ee.lbl.gov>
23 ftp://ftp.ee.lbl.gov/tcpdump.tar.Z (3.4)
24
25This directory contains source code for tcpdump, a tool for network
26monitoring and data acquisition. This software was originally
27developed by the Network Research Group at the Lawrence Berkeley
28National Laboratory. The original distribution is available via
29anonymous ftp to ftp.ee.lbl.gov, in tcpdump.tar.Z. More recent
30development is performed at tcpdump.org, http://www.tcpdump.org/
31
32Tcpdump uses libpcap, a system-independent interface for user-level
33packet capture. Before building tcpdump, you must first retrieve and
34build libpcap, also originally from LBL and now being maintained by
35tcpdump.org; see http://www.tcpdump.org/ .
36
37Once libpcap is built (either install it or make sure it's in
38../libpcap), you can build tcpdump using the procedure in the INSTALL
39file.
40
41The program is loosely based on SMI's "etherfind" although none of the
42etherfind code remains. It was originally written by Van Jacobson as
43part of an ongoing research project to investigate and improve tcp and
44internet gateway performance. The parts of the program originally
45taken from Sun's etherfind were later re-written by Steven McCanne of
46LBL. To insure that there would be no vestige of proprietary code in
47tcpdump, Steve wrote these pieces from the specification given by the
48manual entry, with no access to the source of tcpdump or etherfind.
49
50Over the past few years, tcpdump has been steadily improved by the
51excellent contributions from the Internet community (just browse
52through the CHANGES file). We are grateful for all the input.
53
54Richard Stevens gives an excellent treatment of the Internet protocols
55in his book ``TCP/IP Illustrated, Volume 1''. If you want to learn more
56about tcpdump and how to interpret its output, pick up this book.
57
58Some tools for viewing and analyzing tcpdump trace files are available
59from the Internet Traffic Archive:
60
61 http://www.acm.org/sigcomm/ITA/
62
63Another tool that tcpdump users might find useful is tcpslice:
64
65 ftp://ftp.ee.lbl.gov/tcpslice.tar.Z
66
67It is a program that can be used to extract portions of tcpdump binary
68trace files. See the above distribution for further details and
69documentation.
70
71Problems, bugs, questions, desirable enhancements, etc. should be sent
72to the address "tcpdump-workers@lists.tcpdump.org". Bugs, support
73requests, and feature requests may also be submitted on the SourceForge
74site for tcpdump at
75
76 http://sourceforge.net/projects/tcpdump/
77
78Source code contributions, etc. should be sent to the email address
79submitted as patches on the SourceForge site for tcpdump.
80
81Current versions can be found at www.tcpdump.org, or the SourceForge
82site for tcpdump.
83
84 - The TCPdump team
85
86original text by: Steve McCanne, Craig Leres, Van Jacobson
87
88-------------------------------------
89This directory also contains some short awk programs intended as
90examples of ways to reduce tcpdump data when you're tracking
91particular network problems:
92
93send-ack.awk
94 Simplifies the tcpdump trace for an ftp (or other unidirectional
95 tcp transfer). Since we assume that one host only sends and
96 the other only acks, all address information is left off and
97 we just note if the packet is a "send" or an "ack".
98
99 There is one output line per line of the original trace.
100 Field 1 is the packet time in decimal seconds, relative
101 to the start of the conversation. Field 2 is delta-time
102 from last packet. Field 3 is packet type/direction.
103 "Send" means data going from sender to receiver, "ack"
104 means an ack going from the receiver to the sender. A
105 preceding "*" indicates that the data is a retransmission.
106 A preceding "-" indicates a hole in the sequence space
107 (i.e., missing packet(s)), a "#" means an odd-size (not max
108 seg size) packet. Field 4 has the packet flags
109 (same format as raw trace). Field 5 is the sequence
110 number (start seq. num for sender, next expected seq number
111 for acks). The number in parens following an ack is
112 the delta-time from the first send of the packet to the
113 ack. A number in parens following a send is the
114 delta-time from the first send of the packet to the
115 current send (on duplicate packets only). Duplicate
116 sends or acks have a number in square brackets showing
117 the number of duplicates so far.
118
119 Here is a short sample from near the start of an ftp:
120 3.00 0.20 send . 512
121 3.20 0.20 ack . 1024 (0.20)
122 3.20 0.00 send P 1024
123 3.40 0.20 ack . 1536 (0.20)
124 3.80 0.40 * send . 0 (3.80) [2]
125 3.82 0.02 * ack . 1536 (0.62) [2]
126 Three seconds into the conversation, bytes 512 through 1023
127 were sent. 200ms later they were acked. Shortly thereafter
128 bytes 1024-1535 were sent and again acked after 200ms.
129 Then, for no apparent reason, 0-511 is retransmitted, 3.8
130 seconds after its initial send (the round trip time for this
131 ftp was 1sec, +-500ms). Since the receiver is expecting
132 1536, 1536 is re-acked when 0 arrives.
133
134packetdat.awk
135 Computes chunk summary data for an ftp (or similar
136 unidirectional tcp transfer). [A "chunk" refers to
137 a chunk of the sequence space -- essentially the packet
138 sequence number divided by the max segment size.]
139
140 A summary line is printed showing the number of chunks,
141 the number of packets it took to send that many chunks
142 (if there are no lost or duplicated packets, the number
143 of packets should equal the number of chunks) and the
144 number of acks.
145
146 Following the summary line is one line of information
147 per chunk. The line contains eight fields:
148 1 - the chunk number
149 2 - the start sequence number for this chunk
150 3 - time of first send
151 4 - time of last send
152 5 - time of first ack
153 6 - time of last ack
154 7 - number of times chunk was sent
155 8 - number of times chunk was acked
156 (all times are in decimal seconds, relative to the start
157 of the conversation.)
158
159 As an example, here is the first part of the output for
160 an ftp trace:
161
162 # 134 chunks. 536 packets sent. 508 acks.
163 1 1 0.00 5.80 0.20 0.20 4 1
164 2 513 0.28 6.20 0.40 0.40 4 1
165 3 1025 1.16 6.32 1.20 1.20 4 1
166 4 1561 1.86 15.00 2.00 2.00 6 1
167 5 2049 2.16 15.44 2.20 2.20 5 1
168 6 2585 2.64 16.44 2.80 2.80 5 1
169 7 3073 3.00 16.66 3.20 3.20 4 1
170 8 3609 3.20 17.24 3.40 5.82 4 11
171 9 4097 6.02 6.58 6.20 6.80 2 5
172
173 This says that 134 chunks were transferred (about 70K
174 since the average packet size was 512 bytes). It took
175 536 packets to transfer the data (i.e., on the average
176 each chunk was transmitted four times). Looking at,
177 say, chunk 4, we see it represents the 512 bytes of
178 sequence space from 1561 to 2048. It was first sent
179 1.86 seconds into the conversation. It was last
180 sent 15 seconds into the conversation and was sent
181 a total of 6 times (i.e., it was retransmitted every
182 2 seconds on the average). It was acked once, 140ms
183 after it first arrived.
184
185stime.awk
186atime.awk
187 Output one line per send or ack, respectively, in the form
188 <time> <seq. number>
189 where <time> is the time in seconds since the start of the
190 transfer and <seq. number> is the sequence number being sent
191 or acked. I typically plot this data looking for suspicious
192 patterns.
193
194
195The problem I was looking at was the bulk-data-transfer
196throughput of medium delay network paths (1-6 sec. round trip
197time) under typical DARPA Internet conditions. The trace of the
198ftp transfer of a large file was used as the raw data source.
199The method was:
200
201 - On a local host (but not the Sun running tcpdump), connect to
202 the remote ftp.
203
204 - On the monitor Sun, start the trace going. E.g.,
205 tcpdump host local-host and remote-host and port ftp-data >tracefile
206
207 - On local, do either a get or put of a large file (~500KB),
208 preferably to the null device (to minimize effects like
209 closing the receive window while waiting for a disk write).
210
211 - When transfer is finished, stop tcpdump. Use awk to make up
212 two files of summary data (maxsize is the maximum packet size,
213 tracedata is the file of tcpdump tracedata):
214 awk -f send-ack.awk packetsize=avgsize tracedata >sa
215 awk -f packetdat.awk packetsize=avgsize tracedata >pd
216
217 - While the summary data files are printing, take a look at
218 how the transfer behaved:
219 awk -f stime.awk tracedata | xgraph
220 (90% of what you learn seems to happen in this step).
221
222 - Do all of the above steps several times, both directions,
223 at different times of day, with different protocol
224 implementations on the other end.
225
226 - Using one of the Unix data analysis packages (in my case,
227 S and Gary Perlman's Unix|Stat), spend a few months staring
228 at the data.
229
230 - Change something in the local protocol implementation and
231 redo the steps above.
232
233 - Once a week, tell your funding agent that you're discovering
234 wonderful things and you'll write up that research report
235 "real soon now".
| 15
16Please submit patches against the master copy to the tcpdump project on
17sourceforge.net.
18
19formerly from Lawrence Berkeley National Laboratory
20 Network Research Group <tcpdump@ee.lbl.gov>
21 ftp://ftp.ee.lbl.gov/tcpdump.tar.Z (3.4)
22
23This directory contains source code for tcpdump, a tool for network
24monitoring and data acquisition. This software was originally
25developed by the Network Research Group at the Lawrence Berkeley
26National Laboratory. The original distribution is available via
27anonymous ftp to ftp.ee.lbl.gov, in tcpdump.tar.Z. More recent
28development is performed at tcpdump.org, http://www.tcpdump.org/
29
30Tcpdump uses libpcap, a system-independent interface for user-level
31packet capture. Before building tcpdump, you must first retrieve and
32build libpcap, also originally from LBL and now being maintained by
33tcpdump.org; see http://www.tcpdump.org/ .
34
35Once libpcap is built (either install it or make sure it's in
36../libpcap), you can build tcpdump using the procedure in the INSTALL
37file.
38
39The program is loosely based on SMI's "etherfind" although none of the
40etherfind code remains. It was originally written by Van Jacobson as
41part of an ongoing research project to investigate and improve tcp and
42internet gateway performance. The parts of the program originally
43taken from Sun's etherfind were later re-written by Steven McCanne of
44LBL. To insure that there would be no vestige of proprietary code in
45tcpdump, Steve wrote these pieces from the specification given by the
46manual entry, with no access to the source of tcpdump or etherfind.
47
48Over the past few years, tcpdump has been steadily improved by the
49excellent contributions from the Internet community (just browse
50through the CHANGES file). We are grateful for all the input.
51
52Richard Stevens gives an excellent treatment of the Internet protocols
53in his book ``TCP/IP Illustrated, Volume 1''. If you want to learn more
54about tcpdump and how to interpret its output, pick up this book.
55
56Some tools for viewing and analyzing tcpdump trace files are available
57from the Internet Traffic Archive:
58
59 http://www.acm.org/sigcomm/ITA/
60
61Another tool that tcpdump users might find useful is tcpslice:
62
63 ftp://ftp.ee.lbl.gov/tcpslice.tar.Z
64
65It is a program that can be used to extract portions of tcpdump binary
66trace files. See the above distribution for further details and
67documentation.
68
69Problems, bugs, questions, desirable enhancements, etc. should be sent
70to the address "tcpdump-workers@lists.tcpdump.org". Bugs, support
71requests, and feature requests may also be submitted on the SourceForge
72site for tcpdump at
73
74 http://sourceforge.net/projects/tcpdump/
75
76Source code contributions, etc. should be sent to the email address
77submitted as patches on the SourceForge site for tcpdump.
78
79Current versions can be found at www.tcpdump.org, or the SourceForge
80site for tcpdump.
81
82 - The TCPdump team
83
84original text by: Steve McCanne, Craig Leres, Van Jacobson
85
86-------------------------------------
87This directory also contains some short awk programs intended as
88examples of ways to reduce tcpdump data when you're tracking
89particular network problems:
90
91send-ack.awk
92 Simplifies the tcpdump trace for an ftp (or other unidirectional
93 tcp transfer). Since we assume that one host only sends and
94 the other only acks, all address information is left off and
95 we just note if the packet is a "send" or an "ack".
96
97 There is one output line per line of the original trace.
98 Field 1 is the packet time in decimal seconds, relative
99 to the start of the conversation. Field 2 is delta-time
100 from last packet. Field 3 is packet type/direction.
101 "Send" means data going from sender to receiver, "ack"
102 means an ack going from the receiver to the sender. A
103 preceding "*" indicates that the data is a retransmission.
104 A preceding "-" indicates a hole in the sequence space
105 (i.e., missing packet(s)), a "#" means an odd-size (not max
106 seg size) packet. Field 4 has the packet flags
107 (same format as raw trace). Field 5 is the sequence
108 number (start seq. num for sender, next expected seq number
109 for acks). The number in parens following an ack is
110 the delta-time from the first send of the packet to the
111 ack. A number in parens following a send is the
112 delta-time from the first send of the packet to the
113 current send (on duplicate packets only). Duplicate
114 sends or acks have a number in square brackets showing
115 the number of duplicates so far.
116
117 Here is a short sample from near the start of an ftp:
118 3.00 0.20 send . 512
119 3.20 0.20 ack . 1024 (0.20)
120 3.20 0.00 send P 1024
121 3.40 0.20 ack . 1536 (0.20)
122 3.80 0.40 * send . 0 (3.80) [2]
123 3.82 0.02 * ack . 1536 (0.62) [2]
124 Three seconds into the conversation, bytes 512 through 1023
125 were sent. 200ms later they were acked. Shortly thereafter
126 bytes 1024-1535 were sent and again acked after 200ms.
127 Then, for no apparent reason, 0-511 is retransmitted, 3.8
128 seconds after its initial send (the round trip time for this
129 ftp was 1sec, +-500ms). Since the receiver is expecting
130 1536, 1536 is re-acked when 0 arrives.
131
132packetdat.awk
133 Computes chunk summary data for an ftp (or similar
134 unidirectional tcp transfer). [A "chunk" refers to
135 a chunk of the sequence space -- essentially the packet
136 sequence number divided by the max segment size.]
137
138 A summary line is printed showing the number of chunks,
139 the number of packets it took to send that many chunks
140 (if there are no lost or duplicated packets, the number
141 of packets should equal the number of chunks) and the
142 number of acks.
143
144 Following the summary line is one line of information
145 per chunk. The line contains eight fields:
146 1 - the chunk number
147 2 - the start sequence number for this chunk
148 3 - time of first send
149 4 - time of last send
150 5 - time of first ack
151 6 - time of last ack
152 7 - number of times chunk was sent
153 8 - number of times chunk was acked
154 (all times are in decimal seconds, relative to the start
155 of the conversation.)
156
157 As an example, here is the first part of the output for
158 an ftp trace:
159
160 # 134 chunks. 536 packets sent. 508 acks.
161 1 1 0.00 5.80 0.20 0.20 4 1
162 2 513 0.28 6.20 0.40 0.40 4 1
163 3 1025 1.16 6.32 1.20 1.20 4 1
164 4 1561 1.86 15.00 2.00 2.00 6 1
165 5 2049 2.16 15.44 2.20 2.20 5 1
166 6 2585 2.64 16.44 2.80 2.80 5 1
167 7 3073 3.00 16.66 3.20 3.20 4 1
168 8 3609 3.20 17.24 3.40 5.82 4 11
169 9 4097 6.02 6.58 6.20 6.80 2 5
170
171 This says that 134 chunks were transferred (about 70K
172 since the average packet size was 512 bytes). It took
173 536 packets to transfer the data (i.e., on the average
174 each chunk was transmitted four times). Looking at,
175 say, chunk 4, we see it represents the 512 bytes of
176 sequence space from 1561 to 2048. It was first sent
177 1.86 seconds into the conversation. It was last
178 sent 15 seconds into the conversation and was sent
179 a total of 6 times (i.e., it was retransmitted every
180 2 seconds on the average). It was acked once, 140ms
181 after it first arrived.
182
183stime.awk
184atime.awk
185 Output one line per send or ack, respectively, in the form
186 <time> <seq. number>
187 where <time> is the time in seconds since the start of the
188 transfer and <seq. number> is the sequence number being sent
189 or acked. I typically plot this data looking for suspicious
190 patterns.
191
192
193The problem I was looking at was the bulk-data-transfer
194throughput of medium delay network paths (1-6 sec. round trip
195time) under typical DARPA Internet conditions. The trace of the
196ftp transfer of a large file was used as the raw data source.
197The method was:
198
199 - On a local host (but not the Sun running tcpdump), connect to
200 the remote ftp.
201
202 - On the monitor Sun, start the trace going. E.g.,
203 tcpdump host local-host and remote-host and port ftp-data >tracefile
204
205 - On local, do either a get or put of a large file (~500KB),
206 preferably to the null device (to minimize effects like
207 closing the receive window while waiting for a disk write).
208
209 - When transfer is finished, stop tcpdump. Use awk to make up
210 two files of summary data (maxsize is the maximum packet size,
211 tracedata is the file of tcpdump tracedata):
212 awk -f send-ack.awk packetsize=avgsize tracedata >sa
213 awk -f packetdat.awk packetsize=avgsize tracedata >pd
214
215 - While the summary data files are printing, take a look at
216 how the transfer behaved:
217 awk -f stime.awk tracedata | xgraph
218 (90% of what you learn seems to happen in this step).
219
220 - Do all of the above steps several times, both directions,
221 at different times of day, with different protocol
222 implementations on the other end.
223
224 - Using one of the Unix data analysis packages (in my case,
225 S and Gary Perlman's Unix|Stat), spend a few months staring
226 at the data.
227
228 - Change something in the local protocol implementation and
229 redo the steps above.
230
231 - Once a week, tell your funding agent that you're discovering
232 wonderful things and you'll write up that research report
233 "real soon now".
|