Description
This assignment develops a tool that can be used to explore Internet and TCP performance. The
initial task is to write a simple Web Client that communicates with any Web Server via TCP.
Implemented on top of TCP, HTTP uses a request/response paradigm. The following is an
example request generated by a real browser for / located on www.cnn.com at port 80. The first
line of the request contains the type of request (you only need to use GET, but other types such
as HEAD and POST are possible in HTTP). Following the GET request is the requested object.
The remainder of the line identifies the HTTP version used by the browser. The remaining lines
are HTTP request headers, which you will not need.
GET / HTTP/1.1
Host: www.cnn.com
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.0.3) Gecko/20060426 Firefox/1.5.0.3
Accept:text/xml,application/xml,application/xhtml+xml,text/html;q=0.9,text/plain;q=0.8,image/png,*/*;q=0.5
Accept-Language: en-us,en;q=0.5
Accept-Encoding: gzip,deflate
Accept-Charset: ISO-8859-1,utf-8;q=0.7,*;q=0.7
Keep-Alive: 300
Connection: keep-alive
Note that the HTTP specification expects all lines to be terminated with a CR (carriage return)
and LF (line feed) characters. These characters are \r\n in C/C++. Note also that the HTTP
specification expects a “blank” line containing only \r\n at the end of the request headers.
There are a number of HTTP response codes, but for this project you will simply save the
response code and the returned content.
Client Testing
Start this assignment by building a simple Web Client. This client should connect to a given port
on a given host and send a minimal request string. You will use command line arguments to
control your client where the first argument is a URL of the form http://server:port/path where
the specification of http:// and the port are optional (80 is the default port). For example
./webclient www.cnn.com/
can be used to request the object from port 80 on the machine www.cnn.com. Your simple web
client needs to connect to the port and send the GET line (ending in CR/LF) along with the
Host: line followed by a blank line with CR/LF. You can use “HTTP/1.0” as the version. Note
– the machine goes on the Host line and the path goes in the GET line in the string buffer sent to
CS3516 Computer Networks A15
2
the server. Your client should then receive back the response headers and content from the
server and save them to the file named web.txt by default. You should be able to use your Web
client with any Web server by sending to the standard Web server port 80. Your program should
support the -f option (specified after the URL) allowing an alternate file name to be specified for
the output. For example
./webclient www.cnn.com/ -f cnn.save
will save the output in the file cnn.save. You should also implement a -nf option that causes the
output to be read, but not saved in a file. The –nf option and the –f option are mutually exclusive
in that only one of them can be invoked per execution. The -nf option is useful in connection
with latter parts of the project. Completing the simple client with this functionality is worth 15
out of the 30 points for the project.
TCP Ping
A common measurement performed on the Internet is a “ping” from one host to another to
determine the round trip time (RTT) between the two hosts. Unix/Linux systems have a ping
command, which uses the ICMP protocol to measure the RTT, but due to security concerns few
servers now respond to ping requests.
As an alternate approach, you can implement your own version of ping by adding a -ping option
to measure the time it takes your web client to perform the TCP connect(). The gettimeofday()
systems call can be used to determine and record the time prior to your call to connect() and to
determine the time after connect() has completed. The difference between these two times is an
estimate of the RTT. You should print the RTT value in milliseconds (you will have to do
conversion from the second and microsecond fields used by gettimeofday() ).
Once the -ping option is implemented, select five Web servers from around the U.S. and five
additional Web servers from around the world (and outside the US). Use the -ping option of
your client to make at least 20 measurements for each server. After gathering this data, construct
a table listing each server tested along with the minimum, maximum, median and mean ping
time measured for each server. Include this table in a brief report submitted with Program 2.
This portion of the project is worth an additional five points.
TCP Packet Patterns
For the next five points of the project, you will examine the approach used by a TCP connection
for transmission of data. The use of this feature is controlled by a -pkt option. When this option
is specified, your client should record the time and number of bytes for each read in the main
loop of your program. These values should be recorded in an array and not immediately output
in order to avoid I/O during data collection. Use of the -pkt option should also automatically
turn on the -nf option.
Once your program has read all data from the socket and closed it, the program should print out
the time and number of bytes for each read. In many cases each read will correspond to the
CS3516 Computer Networks A15
3
reception of a single packet so that the number of bytes read is the size of a packet. In other
cases, a read might contain the data of multiple packets. Your program should be able to read up
to 10K bytes on each iteration.
Your report should print the maximum packet size for each of the servers in your test set. Is the
size the same for all servers?
TCP transmission behavior, which we will discuss in more detail later in the course, often results
in bursts of packets being sent. Each of these bursts is sometimes referred to as a “flight.” To
better see this behavior select an object download from a server that requires 30-50 packets. For
three such downloads create a graph where the x-axis is time (time begins just prior to the
connect() call) and the y-axis is the cumulative number of bytes that have been read by the
client. How many packets are read in each flight for different servers? You should include
graphs and analysis in the report you submit. How well formed are these flights in terms of
distinct waves being visually evident in your graphs?
TCP Information
For the final five points on the project, you need to add an -info option to your client, which
displays information about the TCP connection. TCP information maintained by the OS kernel
is available on Linux systems via the TCP_INFO option to the getsockopt() system call. This
information is returned in the tcp_info structure that is defined in the <netinet/tcp.h> include
file. This structure contains much information about the TCP connection, but for this assignment
retrieve and print out the RTT and RTT variance for the TCP connection after all data has been
read from the socket, but before the socket is closed. Include these measured values in your
report. Compare these measured RTT values with the results from the -ping option.
What to turnin for Program 2
Turn in your assignment using the turnin program. You should turn in a tarred file that
includes: the final source program webclient.c , a README file and a make file. The final
version of your Web Client should contain the code for all the implemented options. You must
submit a separate hard copy of your report to Professor Kinicki by the 4 p.m. deadline.
