Assignment 2

Getting started

Please clone the git repository from Assignment 2 and work directly from it so that you can easily pull changes from GitHub should any be necessary.

> git clone https://github.com/softeng364/assignment2.git
> cd assignment2

Submission & Assessment

• A marking rubric is available on Canvas.

• Please upload your code to Canvas.

• The rubric includes a Best Practice component. Fur full credit, please ensure that your code adheres to the usual guidelines of consistency and readability, and use pycodestyle to check conformance with PEP 8, as demonstrated in lab worksheet.

---

Part 1. Link-state routing

Task 1.1. Predecessor lists

Modify dijkstra_5_14() so as to return predecessors (in addition to distances).

Using your new function, visualize the least-cost path tree for node u on the network from Slide 5-15, as we did in Lab 1.

• The new interface should matches that of NetworkX's dijkstra_predecessor_and_distance() i.e. having the same order- and types of output arguments.

• We need to initialize the predecessor map appropriately and update it each time the distance map is update. Only a few lines of code are required.

Task 1.2. Forwarding tables

Write a function forwarding that produces the forwarding table associated with a predecessor map.

• Verify that the output is consistent with the example on Slide 5-17.

• Nodes may be visited more than once. If you have time and would like to experiment with a more efficient algorithm, please see Assignment 2 Challenges, below.

Task 1.3. Widest path routing

Make the necessary modifications to dijkstra_5_14 (from Lab 1) to solve instances of the widest path problem.

Add suitable keyword arguments plus, less, infinity, and min_ to the parameter list so that the same code can solve shortest- or widest (or some other similarly structured) path problems based on the operator functions passed into it.

Visualize the widest path tree for node u on the network from Slide 5-15, as we did in Lab 1, interpreting the link 'cost' as bandwidth.

• In Lab 1, we saw that min used a keyword argument used to extend the applicability of Python's built-in function min(..., key=...).

• If none of these arguments are specified, the defaults should correspond to the least-cost path problem.

• Operators like + and < can be represented as functions using lambda expressions (e.g. lambda a, b: a + b) or the definitions provided in module operator (e.g. operator.add).

Additional remarks (17/05) on Task 1.3

• You may like to introduce an additional keyword parameter sourcedist=0 in dijkstra_generalized, as discussed on this Piazza post. This is not essential for Assignment 2, but please do ensure that you understand the role that this parameter should play.

• To visualize the least-cost path and widest path trees on separate plots, the following code should be used:

import matplotlib.pyplot as plt

fig1 = plt.figure(1)
fig1.suptitle('Shortest paths')
# Least-cost path visualization here
# ...
fig1.show()

fig2 = plt.figure(2)
fig2.suptitle('Widest paths')
# Widest path visualization here
# ...
fig2.show()

---

Part 2. Error detection

Task 2.1. One's complement in Python

Write a function hextet_complement(x) to compute the one's complement of a Python int regarded as a fixed-width hextet (16 bits, two bytes/octets, four nibbles).

• Use the the invert operator ~ and a suitable mask, as discussed in the lab.

• Don't worry about handling the case where the argument itself occupies more than one hextet.

Task 2.2. The Internet Checksum

Implement the Internet Checksum in Python.

• Use hextet_complement() to compute the one's complement.

• Your function should work for any Python sequence whose elements are bytes.

• Ensure that you can reproduce the calculation given in Section 3 of IETF 1071.

We'll use this function to check IP packets in Part 3.

Additional remarks (17/05) on Task 2.2

• Please remember (from the lab worksheet) that an implementation in C of the Internet Checksum is provided in RFC 1071, Section 4.

/*
The following "C" code algorithm computes the checksum with an inner loop that sums 16-bits at a time in a 32-bit accumulator.
*/
{
    /*
    Compute Internet Checksum for "count" bytes beginning at location "addr".
    */
    register long sum = 0;

    while( count > 1 )  {
    /*  This is the inner loop */
        sum += * (unsigned short) addr++;
        count -= 2;
    }

    /*  Add left-over byte, if any */
    if( count > 0 )
        sum += * (unsigned char *) addr;

    /*  Fold 32-bit sum to 16 bits */
    while (sum>>16)
        sum = (sum & 0xffff) + (sum >> 16);

    checksum = ~sum;
}

• With respect to the C code above:

  • A literal translation into Python of the C code sum += * (unsigned short) addr++; is not possible. Rather, we will need to treat the two adjacent bytes separately and shift the bits in the "larger" one.
  • Warning: Recall from the worksheet that Python's operator ~ does not work in the same was as C's version.

• If you like, it is possible to index every second element of a sequence data as follows:

>>> data = b'abcdefg'
>>> data[0::2]
b'aceg'
>>> data[1::2]
b'bdf'
>>>

• If you like, Python does have a built-in sum function (cf. this Piazza post).

Task 2.3. Cyclic Redundancy Checks

Implement a function to perform CRC checks with a given generator on an arbitrary sequence of bytes.

• Verify that your function reproduces the calculation of slide 6-15.

• There is no need to store the quotient.

Additional remarks (17/05) on Task 2.3

• Please use the temlate file crc.py, which contains several test cases. If you have already done this task, please rename your existing file crc.py and ensure that your function crc.crc() has the same interface as the one in the template provided; Thank You.

• The sample code provided in the lab worksheet performs one step of the long division; please ensure that you are happy with this. Hence, we need iterate through a sequence of such steps: Very little new code is required because the polynomial coefficients are either 0 or 1.

---

Part 3. ICMP and socket programming

Use your modules icmp and checksum to reimplement ping in Python.

• To send ICMP messages, we'll need to use a so-called raw socket, as returned by the following snippet:

import socket
socket.socket(family=socket.AF_INET,
              type=socket.SOCK_RAW,
              proto=socket.getprotobyname("icmp"))

Your system might require Administrator permissions to open raw sockets: In particular, you may not be able to complete this task on the Faculty's lab PCs.

• Use Python's argparse module to process the following command-line options:

Name	Abbreviation	Type	Default	Help
--timeout	-w	int	1000	Timeout to wait for each reply (milliseconds)
--count	-c	int	4	Number of echo requests to send
hosts		str		URL or IPv4 address of target host(s)

A demonstration of the required line interface is shown in the examples below.

• Use Python's struct and collections.namedtuple libraries to to de/serialize ICMP messages to/from byte sequences.

• Use Python's with statement to guarantee that your socket is closed gracefully. socket module's documentation provides several examples.

• Use a suitable function from the time module to estimate round-trip time (elapsed time).

• Each ICMP echo request should carry the time instant at which it was created/sent. In reality, this would not be necessary, but it relates to one of the challenge problems.

• Use exceptions to signal checksum errors and timeouts. All exceptions should be handled in verbose_ping().

• Use built-in functions to calculate the minimum, maximum, and mean of the calculated round-trip times, as demonstrated in the example below.

• Complete details about the ICMP messages used for ping implementations are detailed on wikipedia.org.

Additional remarks (17/05) on Part 3

• A more detailed template is now (17/05) available to help with Part 3. Please study the new template and replace each "TODO" with suitable code.

• The type specifiers passed to struct.pack and struct.unpack must be consistent with the ICMP protocol's packet format:

Field	Integer Type	Size in bytes
type	unsigned char	1
code	unsigned char	1
checksum	unsigned short	2
identifier	unsigned short	2
sequence_number	unsigned char unsigned short	1 2

• The type of our payload will be float:

>>> import time
>>> type(time.process_time())
<class 'float'>
>>> type(time.perf_counter())
<class 'float'>
>>> type(time.clock())
<class 'float'>

• You can execute ping.py from the IPython console inside Anaconda using %run e.g.

%run ping --help
%run ping --count 5 www.google.com

Example: Command line interface

> python ping.py --help
usage: ping.py [-h] [-w milliseconds] [-c num] host [host ...]

Test a host.

positional arguments:
  host                  URL or IPv4 address of target host(s).

optional arguments:
  -h, --help            show this help message and exit
  -w milliseconds, --timeout milliseconds
                        Timeout to wait for each reply (milliseconds).
  -c num, --count num   Number of echo requests to send.

Example: Successful invocation

>python ping.py www.python.org --count 3
Contacting www.python.org with 36 bytes of data
Reply from 151.101.0.223 in 5ms: ICMPMessage(type=0, code=0, checksum=48791, identifier=33540, sequence_number=0)
Reply from 151.101.0.223 in 12ms: ICMPMessage(type=0, code=0, checksum=35850, identifier=33540, sequence_number=1)
Reply from 151.101.0.223 in 6ms: ICMPMessage(type=0, code=0, checksum=61385, identifier=33540, sequence_number=2)
Ping statistics for 151.101.0.223:
    Packets: Sent = 3, Received = 3, Lost = 0 (0% loss)
Approximate round trip times in milli-seconds:
    Minimum = 5ms, Maximum = 12ms, Average = 7ms

Example: Timeout

Auckland University's web-server doesn't respond to ICMP echo requests:

> python ping.py www.auckland.ac.nz --count 3 --timeout 1500
Contacting www.auckland.ac.nz with 36 bytes of data
Request timed out after 1500ms
Request timed out after 1500ms
Request timed out after 1500ms
Ping statistics for 130.216.159.127:
    Packets: Sent = 3, Received = 0, Lost = 3 (100.0% loss)

• Please be aware of the marking scheme on Canvas - especially the link to view longer description for each task.

---

Network programming challenges

These are not assessed: Please don't look at these unless you have spare time.

Optimized routing functions

Using only the Python standard library, optimize (for speed) your functions.

• Incorporate a priority queue to improve the efficiency of Dijkstra's algorithm. An implementation is available in the Standard Library's heapq and queue modules.

• Re-implement forwarding using a graph traversal algorithm that visits each node only once. The Boost Graph Library is a good place to look for inspiration e.g. depth_first_search

• Use one of NetworkX's graph generators to generate a large test network.

• Use IPython's %time or %timeit magic functions to compare execution times.

Asynchronous ping

Our current implementation waits to receive the current echo response before sending the next echo request. Employ gevent to write an asynchronous version of your ping code, which doesn't block on echo receives.