Lab TCP
Lab TCP
Objective
To see the details of TCP (Transmission Control Protocol). TCP is the main transport layer protocol used
in the Internet.
1
Step 2: Inspect the Trace
Select a long packet anywhere in the middle of your trace whose protocol is listed as TCP. Expand the TCP
protocol section in the middle panel (by using the “+” expander or icon). All packets except the initial
HTTP GET and last packet of the HTTP response should be listed as TCP. Picking a long packet ensures
that we are looking at a download packet from the server to your computer. Looking at the protocol lay-
ers, you should see an IP block before the TCP block. This is because the TCP segment is carried in an IP.
We have shown the TCP block expanded in our figure.
As well as the above fields, there may be other informational lines that Wireshark provides to help you
interpret the packet. We have covered only the fields that are carried across the network.
2
Step 3: TCP Segment Structure
This drawing differs from the text drawing in the book in only minor respects:
• The Header length and Flags fields are combined into a 2-byte quantity. It is not easy to deter-
mine their bit lengths with Wireshark.
• The Urgent Pointer field is shown as dotted. This field is typically not used, and so does not show
up in Wireshark and we do not expect you to have it in your drawing. You can notice its exist-
ence in Wireshark, however, by observing the zero bytes in the segment that are skipped over
as you select the different fields.
• The Options field is shown dotted, as it may or may not be present for the segments in your
trace. Most often it will be present, and when it is then its length will be a multiple of four bytes.
• The Payload is optional. It is present for the segment you viewed, but not present on an Ack-
only segment, for example.
• Note, you can work out sizes yourself by clicking on a protocol block in the middle panel (the
block itself, not the “+” expander). Wireshark will highlight the corresponding bytes in the packet
in the lower panel, and display the length at the bottom of the window. You may also use the
overall packet size shown in the Length column or Frame detail block. See below where a TCP
packet of length 66 is highlighted.
3
Step 4: TCP Connection Setup/Teardown
Three-Way Handshake
To see the “three way handshake” in action, look for a TCP segment with the SYN flag on. These are up at
the beginning of your trace, and the packets that follow it (see below).
The SYN flag is noted in the Info column. You can also search for packets with the SYN flag on using the
filter expression “tcp.flags.syn==1”. (See below)
A “SYN packet” is the start of the three-way handshake. In this case it will be sent from your computer
to the remote server. The remote server should reply with a TCP segment with the SYN and ACK flags
set, or a “SYN ACK packet”. On receiving this segment, your computer will ACK it, consider the connec-
tion set up, and begin sending data, which in this case will be the HTTP request.
4
Step 5: TCP Connection Setup/Teardown
Next, we wish to clear the display filter tcp.flags.syn==1 so that we can once again see all the packets in
our original trace. Do this by clearing the display filter as shown below.
Figure 11: Clearing the display filter TCP segment with SYN flag on
If you do this correctly, you should see the full trace. We are most interested in the first three packets.
5
Below is a time sequence diagram of the three-way handshake in your trace, up to and including the first
data packet (the HTTP GET request) sent by ‘your computer’ when the connection is established. As
usual, time runs down the page, and lines across the page indicate segments.
Figure 13: Time sequence diagram for the TCP three-way handshake
6
Step 5: Connection Options
As well as setting up a connection, the TCP SYN packets negotiate parameters between the two ends
using Options. Each end describes its capabilities, if any, to the other end by including the appropriate
Options on its SYN. Often both ends must support the behavior for it to be used during data transfer.
Common Options include Maximum Segment Size (MSS) to tell the other side the largest segment that
can be received, and Timestamps to include information on segments for estimating the round trip time.
There are also Options such as NOP (No-operation) and End of Option list that serve to format the Op-
tions but do not advertise capabilities. You do not need to include these formatting options in your an-
swer above. Options can also be carried on regular segments after the connection is set up when they
play a role in data transfer. This depends on the Option. For example: the MSS option is not carried on
each packet because it does not convey new information; timestamps may be included on each packet
to keep a fresh estimate of the RTT; and options such as SACK (Selective Acknowledgments) are used
only when data is received out of order.
Our TCP Options are Maximum Segment Size, Window Scale, SACK permitted, and Timestamps. Each of
these Options is used in both directions. There are also the NOP & End of Option List formatting options.
Points to note:
• The teardown is initiated by the computer; it might also be initiated by the server.
• Like the SYN, the FIN flag occupies one sequence number. Thus, when the sequence number of
the FIN is 192, the corresponding Ack number is 193.
• Your sequence numbers will vary. Our numbers are relative (as computed by Wireshark) but
clearly depend on the resource that is fetched. You can tell that it is around 1 MB long.
• The RTT in the FIN exchange is like that in the SYN exchange, as it should be. Your RTT will vary
depending on the distance between the computer and server as before.
7
Step 6: FIN/RST Teardown
Finally, the TCP connection is taken down after the download is complete. This is typically done with FIN
(Finalize) segments. Each side sends a FIN to the other and acknowledges the FIN they receive; it is simi-
lar to the three-way handshake. Alternatively, the connection may be torn down abruptly when one end
sends a RST (Reset). This packet does not need to be acknowledged by the other side.
Below is a picture of the teardown in your trace, starting from when the first FIN or RST is issued until
the connection is complete. It shows the sequence and ACK numbers on each segment.
Points to note:
• The teardown is initiated by the computer; it might also be initiated by the server.
• The teardown is abrupt – a single RST in this case, and then it is closed, which the other end
must accommodate.
• The sequence and Ack numbers do not really matter here. They are simply the (relative
Wireshark) values at the end of the connection.
8
Step 7: TCP Data Transfer
For this part, we are going to launch an older version of Wireshark called Wireshark legacy. You do this
by selecting the Wireshark legacy application as follows.
When it launches, you should open the trace-tcp file which is in your downloads folder from earlier.
9
You should then be presented with the same trace-tcp as used previously in this exercise.
The middle portion of the TCP connection is the data transfer, or download, in our trace. This is the main
event. To get an overall sense of it, we will first look at the download rate over time.
Under the Statistics menu select an “IO Graph” (as shown below).
You should end up with a graph like below. By default, this graph shows the rate of packets over time.
You might be tempted to use the “TCP Stream Graph” tools under the Statistics menu instead. However,
these tools are not useful for our case because they assume the trace is taken near the computer send-
ing the data; our trace is taken near the computer receiving the data.
10
Now we will tweak it to show the download rate with the changes given below
• On the x-axis, adjust the tick interval and pixels per tick. The tick interval should be small enough
to see into the behavior over the trace, and not so small that there is no averaging. 0.1 seconds
is a good choice for a several second trace. The pixels per tick can be adjusted to make the
graph wider or narrower to fill the window. Make this 10. See below.
• On the y-axis, change the unit to be Bits/Tick. The default is Packet/Tick. By changing it, we can
easily work out the bits/sec throughput by taking the y-axis value and scaling as appropriate,
e.g., 10X for ticks of 0.1 seconds.
• Add a filter expression to see only the download packets. So far we are looking at all of the pack-
ets. Assuming the download is from the usual web server port of 80, you can filter for it with a
filter of “tcp.srcport==80”. Don’t forget to press Enter, and you may need to click the
“Graph” button to cause it to redisplay.
• To see the corresponding graph for the upload traffic, enter a second filter in the next box. Again
assuming the usual web server port, the filter is “tcp.dstport==80”. After you press Enter
and click the Graph button, you should have two lines on the graph.
Our graph for this procedure is shown in the figure on next page. From it we can see the sample down-
load rate quickly increase from zero to a steady rate, with a bit of an exponential curve. This is slow-
start. The download rate when the connection is running is approximately 2.5 Mbps. The upload rate is a
steady, small trickle of ACK traffic. Our download also proceeds fairly steadily until it is done. This is the
ideal, but many downloads may display more variable behavior if, for example, the available bandwidth
varies due to competing downloads, the download rate is set by the server rather than the network, or
enough packets are lost to disrupt the transfer.
Note, you can click on the graph to be taken to the nearest point in the trace if there is a feature you
would like to investigate.
Try clicking on parts of the graph and watch where you are taken in the Wireshark trace window.
11
Figure 16: TCP download rate over time via an IO graph
Inspect the packets in the download in the middle of your trace for these features:
• You should see a pattern of TCP segments received carrying data and ACKs sent back to the
server. Typically, there will be one ACK every couple of packets. These ACKs are called Delayed
ACKs. By delaying for a short while, the number of ACKs is halved.
• Since this is a download, the sequence number of received segments will increase; the ACK
number of subsequently transmitted segments will increase correspondingly.
• Since this is a download, the sequence number of transmitted segments will not increase (after
the initial get). Thus the ACK number on received segments will not increase either.
• Each segment carries Window information to tell the other end how much space remains in the
buffer. The Window must be greater than zero, or the connection will be stalled by flow control.
Note the data rate in the download direction in packets/second and bits/second once the TCP connec-
tion is running well is 250 packet/sec and 2.5 Mbps.
Our download packets are 1434 bytes long, of which 1368 bytes are the TCP payload carrying contents.
Thus 95% of the download is content.
The data rate in the upload direction in packets/second and bits/second due to the ACK packets is 120
packets/sec and 60,000 bits/sec. We expect the ACK packet rate to be around half of the data packet
rate for the typical pattern of one delayed ACK per two data packets received. The ACK bit rate will be
at least an order of magnitude below the data bit rate as the packets are much smaller, around 60 bytes.
The Ack number tells the next expected sequence number therefore it will be X plus the number of TCP
payload bytes in the data segment.
12