> 科技

代写COMP3331/9331 Computer Networks

科技 2024-04-04 08:56:55


COMP3331/9331 Computer Networks and Applications
Assignment for Term 1, 2024
Version 1.0
Due: 11:59am (noon) Thursday, 18 April 2024 (Week 10)
1. Change Log
Version 1.0 released on 7th March 2024.
2. Goal and learning objectives
For this assignment, you are to implement a reliable transport protocol over the UDP protocol. We
will refer to the reliable transport protocol that you will be implementing in this assignment as
Simple Transport Protocol (STP). STP will include most (but not all) of the features that are
described in Sections 3.5.4 - 3.5.6 of the text Computer Networking by Kurose and Ross (7th or 8th
ed.) or equivalent parts from the Week 4/5 lecture notes. Examples of these features include
timeout, ACK, sequence numbers, sliding window, etc. Note that these features are commonly
found in many transport protocols. Therefore, this assignment will give you an opportunity to
implement some of these basic features of a transport protocol. In addition, you may have wondered
why the designer of the TCP/IP protocol stack includes such a feature-less transport protocol as
UDP. You will find in this assignment that you can design your own transport protocol and run it
over UDP. This is the case for some multimedia delivery services on the Internet, where they have
implemented their own proprietary transport protocol over UDP. QUIC, a newly proposed transport
protocol also runs over UDP and implements additional functionalities such as reliability.
Recall that UDP provides point-to-point, unreliable datagram service between a pair of hosts. In this
programming assignment, you will develop a more structured protocol, STP, which ensures
reliable, end-to-end delivery of data in the face of packet loss. STP provides a byte-stream
abstraction like TCP and sends pipelined data segments using a sliding window. However, STP
does not implement congestion control or flow control. Finally, whereas TCP allows fully
bidirectional communication, your implementation of STP will be asymmetric. There will be two
distinct STP endpoints, "sender" and "receiver", respectively. Data packets will only flow in the
"forward" direction from the sender to the receiver, while acknowledgments will only flow in the
"reverse" direction from the receiver back to the sender. To support reliability in a protocol like
STP, state must be maintained at both endpoints. Thus, as in TCP, connection set-up and connection
teardown phases will be an integral part of the protocol. STP should implement a sliding window
protocol wherein multiple segments can be sent by the sender in a pipelined manner. Like TCP,
STP will include some elements of both Go-Back-N (GBN) and Selective Repeat (SR). You will
use your STP protocol to transfer a text file (examples provided on the assignment webpage) from
the sender to the receiver.
The sender program must also emulate the behaviour of an unreliable communication channel
between the sender and receiver. Even though UDP segments can get lost, the likelihood of such
losses is virtually zero in our test environment, where the sender and receiver will be executed on
the same machine. Further, to properly test the implementation of your sender program, we would
like to control the unreliable behaviour of the underlying channel. The sender program should
Updates to the assignment, including any corrections and clarifications, will be posted on the
course website. Please make sure that you check the course website regularly for updates.2
emulate loss of STP segments in both directions – (i) DATA, SYN, and FIN segments in the
forward direction and (ii) ACK segments in the reverse direction. You may assume that the
underlying channel will never reorder or corrupt STP segments (in both directions).
Note that it is mandatory that you implement STP over UDP. Do not use TCP sockets. You
will not receive any marks for this assignment if you use TCP sockets.
2.1 Learning Objectives
On completing this assignment, you will gain sufficient expertise in the following skills:
1. Detailed understanding of how reliable transport protocols such as TCP function.
2. Socket programming for UDP transport protocol.
3. Protocol and message design.
Non-CSE Student Version: The rationale for this option is that students enrolled in a program that
does not include a computer science component have had very limited exposure to programming
and in particular working on complex programming assignments. A Non-CSE student is a student
who is not enrolled in a CSE program (single or double degree). Examples would include students
enrolled exclusively in a single degree program such as Mechatronics or Aerospace or Actuarial
Studies or Law. Students enrolled in dual degree programs that include a CSE program as one
of the degrees do not qualify. Any student who meets this criterion and wishes to avail of this
option MUST email cs3331@cse.unsw.edu.au to seek approval before 5pm, 29th March (Friday,
Week 7). If approved, we will send you the specification for the non-CSE version of the
assignment. We will assume by default that all students are attempting the CSE version of the
assignment unless they have sought explicit permission. No exceptions.
3. Assignment Specification
STP should be implemented as two separate programs: sender and receiver. You should implement
unidirectional transfer of data from the sender to the receiver. As illustrated in Figure 1, data
segments will flow from sender to receiver while ACK segments will flow from receiver to sender.
The sender and receiver programs will be run from different terminals on the same machine, so you
can use localhost, i.e., 127.0.0.1 as the IP address for the sender and receiver in your program. Let
us reiterate this, STP must be implemented on top of UDP. Do not use TCP sockets. If you use
TCP, you will not receive any marks for your assignment.
You will find it useful to review Sections 3.5.4 - 3.5.6 of the text (or the relevant parts from the
Week 5 lecture notes). It may also be useful to review the basic concepts of reliable data transfer
from Section 3.4 (or relevant parts from the Week 4 lecture notes). Section 3.5 of the textbook
which covers the bulk of the discussion on TCP is available to download on the assignment page.
Figure 1: This depicts the assignment setup. A file is to be transferred from the Sender to the Receiver, both running on the same
machine. Data segments will flow from the sender to receiver, while ACK segments will flow from the receiver to sender.
Data
Ack
Sender
Receiver
UDP Socket1
sender_port specified as argument
UDP Socket 2
receiver_port specified as argument3
3.1 File Names
The main code for the sender should be contained in the following files: sender.c, or
Sender.java, or sender.py. You may create additional files such as header files or other
class files and name them as you wish.
The sender should accept the following seven arguments:
1. sender_port: the UDP port number to be used by the sender to send STP segments to the
receiver. The sender will receive ACK segments from the receiver through this same port. We
recommend using a random port number between 49152 to 65535 (dynamic port number
range) for the sender and receiver ports.
2. receiver_port: the UDP port number on which receiver is expecting to receive STP
segments from the sender. The receiver should send ACK segments to the sender through this
same port. We recommend using a random port number in the same range noted above.
3. txt_file_to_send: the name of the text file that must be transferred from sender to
receiver using your reliable transport protocol. You may assume that the file included in the
argument will be available in the current working directory of the sender with the “read”
access permissions set (execute “chmod +r txt_file_to_send” at the terminal in the
directory containing the file, where txt_file_to_send is set to the actual name of the
file).
4. max_win: the maximum window size in bytes for the sender window. This should be an
unsigned integer. Effectively, this is the maximum number of data bytes that the sender can
transmit in a pipelined manner and for which ACKs are outstanding. max_win must be
greater than or equal to 1000 bytes (MSS) and does not include STP headers. When
max_win is set to 1000 bytes, STP will effectively behave as a stop-and-wait protocol,
wherein the sender transmits one data segment at any given time and waits for the
corresponding ACK segment. While testing, we will ensure that max_win is a multiple of
1000 bytes (e.g., 5000 bytes).
5. rto: the value of the retransmission timer in milliseconds. This should be an unsigned
integer.
6. flp: forward loss probability, which is the probability that any segment in the forward
direction (DATA, FIN, SYN) is lost. This should be a float value between 0 and 1 (inclusive).
If flp is 0.1, then the sender will drop approximately 10% of the segments that it intends to
send to the receiver.
7. rlp: reverse loss probability, which is the probability of a segment in the reverse direction
(i.e., ACKs) being lost. This should be a float value between 0 and 1 (inclusive). If rlp is
0.05, then the sender will drop approximately 5% of the ACK segments received.
The sender should be initiated as follows:
If you use Java:
java Sender sender_port receiver_port txt_file_to_send max_win rto flp rlp
If you use C:
./sender sender_port receiver_port txt_file_to_send max_win rto flp rlp
If you use Python:
python3 sender.py sender_port receiver_port txt_file_to_send max_win rto flp rlp4
During testing, we will ensure that the 7 arguments provided are in the correct format. We will not
test for erroneous arguments, missing arguments, etc. That said, it is good programming practice to
check for such input errors.
The main code for the receiver should be contained in the following files: receiver.c, or
Receiver.java, or receiver.py. You may create additional files such as header files or
other class files and name them as you wish.
The receiver should accept the following four arguments:
1. receiver_port: the UDP port number to be used by the receiver to receive STP segments
from the sender. This argument should match the second argument for the sender.
2. sender_port: the UDP port number to be used by the sender to send STP segments to the
receiver. This argument should match the first argument for the sender.
3. txt_file_received: the name of the text file into which the data sent by the sender
should be stored (this is the file that is being transferred from sender to receiver). You may
assume that the receiver program will have permission to create files in its working directory
(execute “chmod +w .” at the terminal to allow the creation of files in the working
directory), and that a file with this name does not exist in the working directory or may be
overwritten.
4. max_win: the receive window size. This argument should match the fourth argument for the
sender. It is provided to the receiver to ensure that the receiver can initialise a buffer
sufficient to hold all the data that the sender can transmit in a pipelined manner.
The receiver should be initiated as follows:
If you use Java:
java Receiver receiver_port sender_port txt_file_received max_win
If you use C:
./receiver receiver_port sender_port txt_file_received max_win
If you use Python:
python3 receiver.py receiver_port sender_port txt_file_received max_win
During testing, we will ensure that the 4 arguments provided are in the correct format. We will not
test for erroneous arguments, missing arguments, etc. That said, it is good programming practice to
check for such input errors.
The receiver must be initiated before initiating the sender. The two programs will be executed on
the same machine. Pay attention to the order of the port numbers to be specified in the arguments
for the two programs as they are in reverse order (sender port is first for the sender while receiver
port is first for the receiver). If you receive an error that one or both port numbers are in use, then
choose different values from the dynamic port number range (49152 to 65535) and try again.
The sender and receiver should exit after the file transfer is complete and the required information,
as stated in the subsequent sections of this document, is written to the sender and receiver log files.
3.2 Segment Format
STP segments must have 2 *two*-byte fields: "type" and "seqno" headers. Each of these store
unsigned integer values in network byte order, i.e., big-endian.
seqno (2 bytes)
Data (0 to MSS bytes)
type (2 bytes)5
The "type" field takes on 4 possible values. DATA = 0, ACK = 1, SYN = 2, FIN = 3.
Unlike TCP, in which multiple types can be set simultaneously, STP segments must be of exactly
one of the types specified above.
The "seqno" field indicates the sequence number of the segment. This field is used in all segments.
For DATA segments, the sequence number increases by the size (in bytes) of each segment
(excluding headers). For ACK segments, the sequence number acts as a cumulative
acknowledgment, and indicates the number of the next byte expected by the receiver. For SYN
segments, the sequence number is the initial sequence number (ISN), which should be a randomly
chosen integer from 0 to 2^16 - 1, which is the maximum sequence number. The sequence number
of the first DATA segment of the connection should thus be ISN+1. For FIN segments, the
sequence number is one larger than the sequence number of the last byte of the last data segment of
the connection. The Maximum Segment Size (MSS) (excluding headers) for STP segment is 1000
bytes. A DATA segment can thus be up to 1004 bytes long. The last DATA segment for the file
being transferred may contain less than 1000 bytes as the file size may not be a multiple of 1000
bytes. All segments excluding DATA segments should only contain the headers and must thus be
exactly 4 bytes long.
The logic for determining the sequence number and ack number in STP is like TCP. However, STP
does not use a separate ack number header field. Rather, the “seqno” field contains the ack number
for the ACK segments.
3.3 State Diagram
The asymmetry between sender and receiver leads to somewhat different state diagrams for the two
endpoints. The state diagram for STP is shown below, which depicts the normal behaviour for both
end points.
Figure 2: State diagram depicting the normal behaviour of STP.
The receiver can be in four possible states: CLOSED, LISTEN, ESTABLISHED and TIME_WAIT.
Initially, it is in the CLOSED state. Upon issuing a passive open, it enters the LISTEN state. Note
that the receiver is the passive host in our protocol and is initiated first, while the sender is initiated
next and actively opens the connection. While in the LISTEN state, the receiver waits for a SYN
segment to arrive on the correct port number. When it does, it responds with an ACK, and moves to
the ESTABLISHED state. Should a duplicate SYN arrive while the receiver is in the
ESTABLISHED state, it reissues the ACK and remains in the ESTABLISHED state. The ACKs
sent by the receiver are cumulative (like TCP). After the sender has reliably transmitted all data 6
(and received acknowledgments), it will send a FIN segment to the receiver. Upon receipt of the
FIN, the receiver moves to the TIME_WAIT state. As in TCP, it remains in TIME_WAIT for two
maximum segment lifetimes (MSLs) before re-entering the CLOSED state. This is to ensure that
the receiver can respond to potentially retransmitted FIN segments from the sender. You may
assume that the MSL is 1 seconds. In other words, the receiver should remain in TIME_WAIT for 2
seconds and then transition to CLOSED.
The sender can be in five possible states: CLOSED, SYN_SENT, ESTABLISHED, CLOSING and
FIN_WAIT. Like the receiver, the sender starts in the CLOSED state. It then issues an active open
by sending a SYN segment (to the receiver's port), thus entering the SYN_SENT state. This SYN
transmission also includes the initial sequence number (ISN) of the conversation. The ISN should
be chosen at random from the valid range of possible sequence numbers (0 to 2^16 – 1). If a
corresponding ACK is not received within rto msec, the sender should retransmit the SYN
segment. In the common case in which the SYN is acknowledged correctly (the ACK must have the
correct sequence number = ISN + 1), the sender enters the ESTABLISHED state and starts
transmitting DATA segments. The sender maintains a single timer (for rto msec) for the oldest
unacknowledged packet and only retransmits this packet if the timer expires. Once all DATA
segments have been transmitted the sender enters the CLOSING state. At this point, the sender must
still ensure that any buffered data arrives at the receiver reliably. Upon verification of successful
transmission, the sender sends a FIN segment with the appropriate sequence number (1 greater than
the sequence number of the last data byte) and enters the FIN_WAIT state. Once the FIN segment
is acknowledged, the sender re-enters the CLOSED state. If an ACK is not received before the
timer (rto msec) expires, the sender should retransmit the FIN segment.
Strictly speaking, you don’t have to implement the CLOSED state at the start for the sender. Your
sender program when executed can immediately send the SYN segment and enter the SYN_SENT
state. Also, when the sender is in the FIN_WAIT state and receives the ACK for the FIN segment,
the program can simply exit. This is because the sender only transmits a single file in one execution
and quits following the reliable file transfer.
Unlike TCP which follows a three-way handshake (SYN, SYN/ACK, ACK) for connection setup
and independent connection closures (FIN, ACK) in each direction, STP follows a two-way
connection setup (SYN, ACK) and one directional connection closure (FIN, ACK) process. The
setup and closure are always initiated by the sender.
If one end point detects behaviour that is unexpected, it should simply print a message to the
terminal indicating that the connection is being reset, and then terminate. For example, if the
receiver receives a DATA segment while it is in the LISTEN state (where it is expecting a SYN
segment). The state transition diagram on the previous page does not capture such erroneous
scenarios. You are free to specify the format of the message that is printed. Note that, we will NOT
be rigorously testing your code for such unexpected behaviour.
3.4 List of features to be implemented by the sender.
You are required to implement the following features in the sender (and equivalent functionality in
the receiver).
1. The sender should first open a UDP socket on sender_port and initiate a two-way
handshake (SYN, ACK) for the connection establishment. The sender sends a SYN segment, and
the receiver responds with an ACK. This is different to the three-way handshake implemented by
TCP. If the ACK is not received before a timeout (rto msec), the sender should retransmit the
SYN.
2. The sender must choose a random initial sequence number (ISN) between 0 and 216 – 1. 7
Remember to perform sequence number arithmetic modulo 216. The sequence numbers should cycle
back to zero after reaching 216 – 1.
3. A one-directional (forward) connection termination (FIN, ACK). The sender will initiate the
connection close once the entire file has been reliably transmitted by sending the FIN segment and
the receiver will respond with an ACK. This is different to the bi-directional close implemented by
TCP. If the ACK is not received before a timeout (rto msec), the sender should retransmit the FIN.
The sender should terminate after connection closure.
4. STP implements a sliding window protocol like TCP, whereby multiple segments can be
transmitted by the sender in a pipelined manner. The sender should maintain a buffer to store all
unacknowledged segments. The total amount of data that the sender can transmit in a pipelined
manner and for which acknowledgments are pending is limited by max_win. Similar to TCP, as
the sender receives ACK segments, the left edge of the window can slide forward, and the sender
can transmit the next DATA segments (if there is pending data to be sent).
5. Each STP segment transmitted by the sender (including DATA, SYN, FIN) must be encapsulated
in a UDP segment and transmitted through the UDP socket.
6. The sender must maintain a single timer for retransmission of data segments (Section 3.5.4 of the
text). The value of the timeout will be supplied as a command-line argument to the sender program
(rto msec). This timer is for the oldest unacknowledged data segment. In the event of a timeout,
only the oldest unacknowledged data segment should be retransmitted (like TCP). The sender
should not retransmit all unacknowledged segments. Remember that you are NOT implementing
Go-Back-N.
7. The sender should implement all the features mentioned in Section 3.5.4 of the text, except for
doubling the timeout interval. You are expected to implement the functionality of the simplified
TCP sender (Figure 3.33 of the text) and fast retransmit (i.e., the sender should retransmit the oldest
unacknowledged data segment on three duplicate ACKs) (pages 247-248).
8. The use of the “seqno” field was outlined in Section 3.2. For DATA segments, the sequence
number increases by the size (in bytes) of each segment (excluding headers). For ACK segments,
the sequence number acts as a cumulative acknowledgment, and indicates the number of the next
byte expected by the receiver. The logic is thus like TCP, except that STP does not use a separate
ACK header field. The ACK segments use the seqno header field to indicate the ACK numbers.
9. The sender will receive each ACK segment from the receiver through the same socket which the
sender uses to transmit data. The ACK segment will be encapsulated in a UDP segment. The sender
must first extract the ACK segment from the UDP segment and then process it as per the operation
of the STP protocol. ACK segments have the same format as DATA segments but do not contain
any data.
10. The sender program should emulate the behaviour of an unreliable communication channel
between the sender and receiver. UDP segments can occasionally experience loss in a network, but
the likelihood is very low when the sender and receiver are executed on the same machine.
Moreover, to properly test the implementation of your sender program, we would like to control the
unreliable behaviour of the underlying channel. The sender emulates loss of STP segments in both
directions which can be controlled through two command line argument: (i) flp: determines the
probability of a DATA, SYN, or FIN segment created by the sender being dropped. In other words,
each such segment created by the sender will be dropped with a probability rlp. If the segment is
not dropped, then it will be transmitted through the socket to the receiver. (ii) rlp: determines the
probability of an ACK segment in the reverse direction from the sender being dropped. In other
words, each ACK segment arriving at the sender socket will be dropped with a probability flp. If
the segment is not dropped, then it will be processed as per the STP protocol.8
Note about Random Number Generation
You will need to generate random numbers to establish an initial sequence number (ISN) and
implement segment loss. If you have not learnt about the principles behind random number
generators, you need to know that random numbers are in fact generated by a deterministic formula
by a computer program. Therefore, strictly speaking, random number generators are called pseudo
random number generators because the numbers are not truly random. The deterministic formula
for random number generation in Python, Java and C use an input parameter called a seed. If a fixed
seed is used, then the same sequence of random numbers will be produced, each time the program is
executed. This will thus likely generate the same ISN and the same sequence of segment loss in
each execution of the sender. While this may be useful for debugging purposes, it is not a realistic
representation of an unreliable channel. Thus, you must ensure that you do not use a fixed seed in
your submitted program. A simple way to use a different seed for each execution is to base the seed
on the system time.
The following code fragments in Python, Java and C generate random numbers between 0 and 1
with a different seed in each execution.

In Python, you initialise a random number generator by using random.seed(). By default,
the random number generator uses the current system time. After that you can generate a
random floating point number between (0,1) by using random.random().

In Java, you initialise a random number generator by using Random random = new
Random(). This constructor sets the seed of the random number generator to a value very
likely to be distinct from any other invocation of this constructor. After that, you can generate a
random floating point number between (0,1) by using
float x =
random.nextFloat().

In C, you initialise a random number generator by using srand(time(NULL)). After that,
you can generate a random floating point number between (0,1) by using float x =
rand()/((float)(RAND_MAX)+1). Note that, RAND_MAX is the maximum value
returned by the rand() function.
11. The sender should maintain a log file named sender_log.txt where it records the
information about each segment that it sends and receives. Information about dropped segments
should also be included. You may assume that the sender program will have permission to create
files in its current working directory. If the file already exists, then it should be overwritten. Start
each entry on a new line. The format should be as follows:
<snd|rcv|drp> <time> <type-of-segment> <seq-number> <number-of-bytes>
There must be no whitespace within each field, and all fields must be whitespace separated. You
may choose how to separate each field, e.g. with padding or a fixed number of spaces/tabs, but the
log file must be easily human readable.
The first field is from the perspective of the sender endpoint, i.e., the sender will snd DATA, FIN,
and SYN segments, if they are not dropped, and rcv ACK segments, if they are not dropped. Any
dropped segments will be marked as drp. The time should be in milliseconds and relative to when
the first SYN segment was sent (even if it was dropped), thus the first entry will be a SYN segment
at time 0. The type of segment can be ACK, DATA, SYN or FIN. The number of bytes should be
zero for all segments other than DATA segments, where it is the number of bytes of actual payload
data (i.e., without headers).9
The log must be recorded in chronological order.
For example, the following shows the log file for a sender that transmits 3000 bytes of data and the
rto is 100 msec. The ISN chosen is 4521 and max_win is 3000 bytes. Notice that the third
DATA segment is dropped and is hence retransmitted after a timeout interval of 100 msec. The
interval is quite precise in this example log, but minor variations due to system latency are expected
and acceptable.
snd 0.00
SYN 4521 0
rcv 10.34
ACK 4522 0
snd 10.45
DATA 4522 1000
snd 10.55
DATA 5522 1000
drp 10.67
DATA 6522 1000
rcv 36.76
ACK 5522 0
rcv 37.87
ACK 6522 0
snd 137.87
DATA 6522 1000
rcv 167.23
ACK 7522 0
snd 168.11
FIN 7522 0
rcv 203.34
ACK 7523 0
Once the entire file has been reliably transmitted and the connection is closed, the sender should
also print the following statistics at the end of the log file (i.e., sender_log.txt). Please note
that these statistics are specific to the sender while it is in an ESTABLISHED or CLOSING state,
i.e., transmitting DATA segments, and receiving the corresponding (and potentially cumulative)
ACK segments. We are not including summary statistics related to SYN or FIN segments and their
acknowledgement. Also note, the amount of data sent and acknowledged is strictly payload data.
1. Amount of (original) Data Sent (in bytes) – regardless of whether the data is dropped but
does not include retransmitted data.
2. Amount of (original) Data Acknowledged (in bytes) – based only on acknowledgements that
are not dropped.
3. Number of (original) Data Segments Sent – regardless of whether the segment is dropped
but does not include retransmitted segments.
4. Number of Retransmitted Data Segments.
5. Number of Duplicate Acknowledgements Received.
6. Number of Data Segments Dropped.
7. Number of Acknowledgements Dropped.
For the example scenario, the sender log would be appended with:
Original data sent:
3000
Original data acked:
3000
Original segments sent:
3
Retransmitted segments:
1
Dup acks received:
0
Data segments dropped:
1
Ack segments dropped:
0
You must follow the above format, but you may choose how to separate each statistical description
with its value, e.g. with padding or a fixed number of spaces/tabs, provided the summary statistics
are easily human readable.
NOTE: Generation of this log file is very important. It will help your tutors in understanding the 10
flow of your implementation and marking. So, if your code does not generate any log files, you will
only be graded out of 25% of the marks.
The sender should terminate after the file transfer is complete and the log file is finalised.
The sender must only print a message to the terminal if it is terminating prematurely, due to
unexpected behaviour by the receiver. No other output should be displayed. If you are printing
output to the terminal for debugging purposes, make sure you disable it prior to submission.
3.5 Specific details about the receiver
1. The receiver should first open a UDP socket on receiver_port and then wait for segments to
arrive from the sender. The first segment to be sent by the sender is a SYN segment and the receiver
will reply with an ACK segment.
2. The receiver should next create a new text file named as per the txt_file_received
argument. You may assume that the receiver program will have permission to create files in its
current working directory (execute “chmod +w .” at the terminal to allow the creation of files in
the working directory). The received data will be written to this file in the correct order.
3. The receiver should initialise a receive window of max_win to hold all data that the sender can
send in a pipelined manner.
4. The receiver should generate an ACK immediately after receiving any segment from the sender.
The receiver should not follow Table 3.2 of the textbook and does not implement delayed ACKs.
The format of the ACK segment is exactly like the STP DATA segment. It should however not
contain any data. The ack number should be included in the “seqno” field of the STP segment.
There is no explicit ACK field in the STP header.
5. The receiver should buffer all out-of-order data in the receive buffer. This is because STP
implements reliable in-order delivery of data.
6. The receiver should write data (in correct order) from the receive buffer to the
txt_file_received file. The file should be written as correctly ordered data is available to
ensure the receiver buffer does not overflow. At the end of the transfer, the receiver should have a
duplicate of the text file sent by the sender. You can verify this by using the diff command on a
Linux machine (e.g., “diff txt_file_to_send txt_file_received”). When testing
your program, if you have the sender and receiver executing in the same working directory then
make sure that the file name provided as the argument to the receiver is different from the file name
used by the sender.
7. Once the file transfer is complete, the receiver should follow the state transition process as
outlined in Section 3.3 while implementing connection closure. Pay particular attention to the
transition from the TIME_WAIT state to the CLOSED state.
8. The receiver should also maintain a log file named receiver_log.txt where it records the
information about each segment that it sends and receives, in chronological order. If the file already
exists, then it should be overwritten. The format should be exactly similar to the sender log file, as
outlined in the sender specification (Section 3.4). The first field is from the perspective of the
receiver endpoint, i.e., the receiver will rcv DATA, FIN, and SYN segments, and snd ACK
segments. Time should be in milliseconds and relative to when the first SYN segment is received,
thus the first entry will be a SYN segment at time 0. One obvious difference from the sender log is
that the receiver log will not record any segments as drp.
For example, the following shows the log file for a receiver that matches the scenario outlined in
the sender specification (Section 3.4). To recall, the sender transmits 3000 bytes of data and the
rto is 100 msec. The ISN is 4521, the max_win is 3000 bytes, and the third DATA segment is 11
dropped, triggering a retransmission after the timeout interval.
rcv 0.00
SYN 4521 0
snd 0.34
ACK 4522 0
rcv 10.65
DATA 4522 1000
snd 10.75
ACK 5522 0
rcv 10.95
DATA 5522 1000
snd 11.03
ACK 6522 0
rcv 142.41
DATA 6522 1000
snd 144.58
ACK 7522 0
rcv 170.45
FIN 7522 0
snd 171.59
ACK 7523 0
Once the entire file has been reliably transmitted and the connection is closed (remember to follow
the state machine for the receiver), the receiver will also print the following statistics at the end of
the log file (i.e., receiver_log.txt). Like the sender log, these statistics are specific to the
receiver while it is in an ESTABLISHED state, i.e., receiving DATA segments, and sending the
corresponding (and potentially cumulative) ACK segments. We are not including summary
statistics related to SYN or FIN segments and their acknowledgement. Also note, the amount of
data received and acknowledged is strictly payload data.
1. Amount of (original) Data Received (in bytes) – does not include retransmitted data.
2. Number of (original) Data Segments Received – does not include retransmitted data.
3. Number of Duplicate Data Segments Received.
4. Number of Duplicate Acknowledgements Sent.
For the example scenario, the receiver log would be appended with:
Original data received:
3000
Original segments received:
3
Dup data segments received:
0
Dup ack segments sent:
0
You must follow the above format, but you may choose how to separate each statistical description
with its value, e.g. with padding or a fixed number of spaces/tabs, provided the summary statistics
are easily human readable.
REMINDER: Generation of this log file is very important. It will help your tutors in understanding
the flow of your implementation and marking. So, if your code does not generate any log files, you
will only be graded out of 25% of the marks.
The receiver should terminate after the file transfer is complete and the log file is finalised.
The receiver must only print a message to the terminal if it is terminating prematurely, due to
unexpected behaviour by the sender. No other output should be displayed. If you are printing output
to the terminal for debugging purposes, make sure you disable it prior to submission.
3.6 Features Excluded
There are several transport layer features adopted by TCP that are excluded from this assignment:
1. You do not need to implement timeout estimation. The timer value is provided as a command
line argument (rto msec).
2. You do not need to double the timeout interval.12
3. You do not need to implement flow control and congestion control.
4. STP does not have to deal with corrupted or reordered segments. Segments will rarely be
corrupted and reordered when the sender and receiver are executing on the same machine. In
short, it is safe for you to assume that packets are only lost. Note however, that segments can
be dropped by the unreliable channel implemented in the sender program and thus segments
may arrive out of order at the receiver.
3.7 Implementation Details
The figure below provides a high-level and simplified view of the assignment. The STP protocol
logic implements the state maintained at the sender and receiver, which includes all the state
variables and buffers. Note that each STP segment (in each direction) must be encapsulated in a
UDP segment and transmitted through the UDP socket at each end point.
Figure 3: The overall structure of your assignment.
Sender Design
The sender must first execute connection setup, followed by data transmission, and finally
connection teardown, as outlined in the state diagram description (Section 3.3). During connection
setup, the sender transmits a SYN segment, starts a timer (rto) and waits for an ACK. During data
transmission, the sender may transmit several STP data segments (determined by max_win), all of
which need to be buffered (in case of retransmissions) and wait for the corresponding ACKs. A
timer (rto) should be started for the oldest unacknowledged segment. During connection
teardown, the sender must transmit a FIN segment, start a timer (rto) and wait for an ACK. Each
STP segment (of any kind) that is to be transmitted must be encapsulated in a UDP segment and
sent through the sender socket to the receiver. The sender should also process incoming ACK
segments from the receiver. Upon receipt of three duplicate ACKs, the sender should retransmit the
oldest unacknowledged DATA segment. In the case of a timeout, the sender should retransmit the
SYN or FIN, if in the connection setup or teardown process, or retransmit the oldest
unacknowledged DATA segment, if in the data transmission process. As the sender needs to deal
with multiple events, we recommend using multi-threading. 13
To better understand the logic behind fast retransmission and the use of a single timer, a simplified
view of the sender is presented below (this does not include the connection setup and teardown and
the functionality for implementing the unreliable channel)
event: data read from text file to be sent
• create STP segment with seqno, where seqno is byte-stream number of first data byte in segment
• transmit segment
• start timer if not already running
• think of timer as for oldest unACKed segment
• expiration interval: rto msec
event: timeout
• retransmit oldest unACKed segment
• restart timer
• reset dupACKcount
event: ACK received
• if ACK acknowledges previously unACKed segment(s)
• update what is known to be ACKed
• if there are currently any unACKed segments
• restart timer
• otherwise (there are no unACKed segments)
• stop timer
• reset dupACKcount
• otherwise (ACK acknowledges previously ACKed segments)
• increment dupACKcount
• if dupACKcount = 3
• retransmit oldest unACKed segment
• reset dupACKcount
Note that the use of a single timer does not require the sender to maintain any additional state for
each segment such as when it was sent.
Recall also that the sender needs to implement the functionality of an unreliable channel that can
drop segments in either direction. Before sending a UDP segment through the socket, the sender
should call the function that simulates segment loss with a probability flp. If the segment is to be
dropped, then it should not be transmitted and nothing else needs to be done (other than updating
the log). Only if the segment is not dropped should it be transmitted through the socket.
Upon receipt of a UDP segment through the socket, the sender should call the function that
simulates packet loss with a probability rlp. If the segment is dropped, then nothing else needs to
be done (other than updating the log). Remember that we are emulating loss of the incoming
segment on the channel, so effectively this segment never arrived at the sender and thus no action is
necessary. If the segment is not dropped, only then should the sender execute the STP protocol
logic (as outlined above and in Sections 3.3, 3.4).
Receiver Design
Upon receipt of a UDP segment through the socket, the receiver should extract the STP segment
which is encapsulated within the UDP segment. The receiver should then execute the STP protocol
logic (as outlined in Sections 3.3 and 3.5). For a SYN or FIN segment, the corresponding ACK
should be sent, and other actions should be taken as per the state diagram shown in Section 3.3. For
a DATA segment, the data should be written to the receive buffer. If the data is received in order, 14
then it can be written to the file and the buffer space made available. If out of order, then it will
remain in the buffer until the missing data is received. An appropriate ACK segment should be
generated. The ACK should then be encapsulated in a UDP segment and sent through the receiver
socket to the sender.
All the above functionalities can be implemented in a single thread. However, the receiver must
implement a timer during connection closure for transitioning from the TIME_WAIT state to the
CLOSED state. The suggested way to implement this is using multi-threading where a separate
thread is used to manage the timer. However, it may be possible to implement this using a single
thread and using non-blocking or asynchronous I/O by using polling, i.e., select().
Data Structures
To manage your protocol, you will need a collection of state variables to help you keep track of
things like state transitions, sliding windows, and buffers. In TCP, this data structure is referred to
as a control block: it's probably a good idea to create a control block class of your own and have a
member variable of this type in your primary class. While we do not mandate the specifics, it is
critical that you invest some time into thinking about the design of your data structures. You should
be particularly careful about how multiple threads will interact with the various data structures.
Note about Artificial Delay
You may be tempted to add artificial delay to your sender and/or receiver through the use of system
calls such as sleep(). For example, to implement your retransmission timer, or to try and resolve
synchronisation issues. You are generally cautioned against doing this. At the very least, it will
introduce unnecessary inefficiency to your protocol. Of perhaps greater concern, if you are doing it
in response to some issue, then most likely the bug will remain. It will just be more transient and
difficult to find. There will almost certainly be a more logical solution that will improve the
efficiency and robustness of your programs. In such scenarios, you are highly encouraged to take a
systematic approach to debugging your code and properly resolve any issues.
4. Test Implementations
Implementations for both endpoints will be made available on the CSE server system. They will be
initiated as follows:
3331 sender sender_port receiver_port txt_file_to_send max_win rto flp rlp
3331 receiver receiver_port sender_port txt_file_received max_win
The command-line arguments are as described in Section 3.1.
Please note that the behaviour of your applications is not expected to match the test
implementations. For example, the test implementations may save files to a temporary directory,
output logs to screen, and provide more explicit details when unexpected behaviour is detected.
You should always defer to the assignment specification as to what is expected of your
implementations.
Recall, the sender and receiver programs will be run from different terminals on the same machine.
For the CSE Linux servers, the easiest way to achieve this is to connect to VLAB and open two
terminals. If you are instead connecting over ssh, then you must be mindful that both connections
are to the same server. Here are two suggested approaches to using ssh:
1. Start a ssh session and determine the hostname from the shell prompt, e.g.,
z1234567@vx01 indicates a connection to host vx01. From another local terminal initiate
a second ssh session. If you don’t connect to the same host then in the second session you
can simply ssh to the first host, e.g., ssh vx01.15
2. Start a ssh session and then use an application like screen to multiplex the one terminal
between both the sender and receiver applications.
If you receive an error when starting the sender or receiver that a port number is in use, then choose
a different value from the dynamic port number range (49152 to 65535) and try again.
5. Additional Notes

请加QQ:99515681  邮箱:99515681@qq.com   WX:codinghelp

Copyright © 2024 东方时讯 All Rights Reserved.

部分内容来源于用户投稿,如果该内容的涉及您的权益,请与我们联系,我们将尽快处理!