CS 276: Distributed Computing and Computer Networks
Homework Assignment #3
1.0 Overview
In this assignment you use unicast TCP connections to simulate a multicast
service over the Internet. The topology of the tree will vary and you will
have to deal with all types of virtual trees. Given a unicast routing topology,
you will be required to build a multicast tree and manage join/leave messages
for all group members. The key performance measure will be whether nodes
implement the multicast routing/forwarding protocol correctly. This will
be determined based on whether each node in the graph handles join/leave
messages correctly and whether each node properly forwards or discards
packets from the source. Your nodes will have to run in both a ``broadcast-and-prune''
and a ``join-and-graft'' mode of operation (but only one protocol at a
time).
You will write a ``node'' program that is run on up to 10 different
machines (though extending beyond 10 should be trivial). Each ``node''
program will read from a topology file and establish a virtual topology
of unicast links. The links will actually be implemented using TCP connections.
Each link will have a virtual cost. Using Node 1 as the one and only source
you will need to build a logical representation of a multicast tree. When
Node 1 begins sending packets, each packet should be delivered across the
multicast tree, but only to nodes that are group members.
Two input files will be needed: the topology file mentioned above,
and an event file. The event file will contain a list of times and corresponding
node actions. These events will be join/leave messages from nodes wishing
to join/leave the multicast group. The node which initiates the event must
send an event message to the next-hop, upstream router who must properly
act on the message, either adding the node to the multicast tree and passing
the message further upstream or pruning the link from the multicast tree.
The output from your programs will be two files for each node.
The first will contain a history of events for each node and the second
will contain a transmission summary. The possible events are for data packets
only and include: transmitted (source node only), forwarded (intermediate
nodes only), discarded, and received.
The transmission summary will include information about the total
number of packets sent, received, forwarded, and dropped by each node.
2.0 Input Given
Two configuration files will be used: (1) one file specifying the topology
of the nodes, and (2) a second file specifying the events that occur during
the simulation.
-
Topology File (click here for an example)
-
You should be able to read the topology configuration file and create a
logical representation of the unicast routing topology in your simulator.
The following format will be used:
<number of nodes in tree> <duration> <interval> <dense/sparse>
<node #> <IP address of host> <starting port number>
<node 1> <node 2> <hop cost>
The dense/sparse option specifies what type of routing protocol will be
run for this particular configuration.
<duration> above is the duration of the simulation
<interval> above is the resolution of time unit used in the simulation
( in seconds ).
Event File (click here for an example)
-
The following format will be used:
<time> <node #> <event type>
<event type> is either join or leave.
The time component will always be an integer greater than zero. The time
component is actually a ``sequence number trigger time''. In addition to
data packets, the source also broadcast heartbeat packets, one every <interval>.
Heartbeat packets are sent to all receivers and are independent of data
packets sent to the multicast address. Heartbeat packets are necessary
in order to provide a coarse-grain global clock to synchronize events.
Note that the heartbeat messages will be sent by source and all the
nodes in the topology will receive it. This transmission will be using
a truncated broadcast tree based. That is, after you create the topology,
you need to create a tree covering all the nodes in the topology. This
tree is then used to transmit heartbeat messages.
3.0 Output Format
Two output files will need to be generated per node:
-
log-<node #>.out: (click here for an example)
-
The log-<node #>.out file should contain a sorted list of events
(you'll have to sort based on the entire line because only the entire line
will be unique) using the following format:
<time> <action node> <secondary node> <action>
The <action> is either join, leave, prune, snd, fwd, rcv, rpf, or
drp.
Most actions involve two nodes: the action node and the secondary node.
The action node is the node in which the event takes place, and the secondary
node (coupled with the action node) is used to identify the link over which
the packet will flow (in the case of a send), was sent (in the case of
a receive), or won't be delivered (in the case of a rpf, ttl, or prn).
-
There are three reasons why a packet may not be forwarded: (1) rpf, reverse
path first check which failed, or (2) prn, dropped because of prune state.
Sometimes there will only be single nodes (in the case of a join or leave).
In this case the secondary node should be made the same as the action node.
summary-<node #>.out: (click here for an
example)
-
The summary-<node #>.out file should contain one line per node,
sorted by node number with the following information:
<node #> <pkts snd> <pkts fwd> <pkts rcv> <pkts rpf> <pkts drp>
4.0 Detailed Operation
The detailed operation is left mainly to the discretion of the student.
However, there are some strict requirements that need to be implemented.
This list is both a list of requirements and a list of suggestions. Be
forewarned that items will be added to the list over the course of the
assignment (as corrections and clarifications are made).
-
For this assignment, there is a strict requirement that you turn in exactly
two files: node.c and a Makefile.
-
The command line parameters for the executable should be:
node <node number> <topology file> <event file>
The two output files created should be log-<node #>.out and
summary-<node
#>.out.
The source will always be Node 1. However, because there is only one node.c
program, the send function will be included in each node. Your node.c
should determine whether it is Node 1 and then initiate the heartbeat and
data packet transmission if it is. Data packets should be sent once every
two intervals, and heartbeat packets should be sent once every interval.
Finally,
ONLY data packets and not heartbeat packets are counted
in the log and summary output files.
Upon startup, each node will need to form a unicast routing table using
a link state style protocol (since all nodes will have access to the topology
and link costs). You will also need to create a truncated broadcast
tree for the heartbeat packets.
-
Upon startup, a TCP connection should be opened for each link specified
in the topology file. The link should be opened by the node with the lowest
number. For this reason, when running a program, always start nodes from
highest to lowest (so that when the lower numbered node in a link is started,
the peer will already exist).
-
Handling port assignments will be a little bit tricky, but use the following
convention: The lower numbered node will connect to the higher numbered
nodes IP address. The port number used will be the number specified in
the topology file plus the node number of the client node. For example,
if 5 is connecting to 6 and 6 has a port number of 38000 in the configuration
file, node 5 should connect to port 38005.
-
You will have to implement a dense-style broadcast-and-prune version and
a sparse-style join-and-graft style of protocol.
-
You will need to create state in each node based on whether traffic should
be forwarded or dropped based on prune state. Since there is only one multicast
group there will only be one entry for each outgoing link.
-
A node should not actually print out any messages while executing. This
is so the nodes can be run in the background, possibly on the same machine.
-
You will need to create both control packets (for joins, leaves, and heartbeat
packets) and data packets. This will require you to create a protocol on
top of TCP so that nodes know whether an incoming packet should be forwarded/dropped
or processed. You are free to implement these headers however you like.
5.0 Turnin
-
Electronic turnin: Use the turnin command to turnin node.c
and Makefile files only.
$> turnin
hw3@cs276 files-and-directories
-
Hardcopy turnin: Turnin a copy of the source code as well as a brief
(1 page max) summary of the differences between sparse and dense mode protocols.
You should identify ways of quantitatively comparing the performance of
the two (packet hops for example) and state whether there are cases when
dense is better than sparse and vice versa. Don't forget to
-
Turnin joint code
-
Turnin separate write-ups
6.0 Grading
We will design several strategies that will test and stress the functionality
of your program. We will develop 4 test cases of increasing complexity.
Each will be tested against dense and sparse mode protocols. Each successful
execution will be worth 5 points.
-
40: dense-mode protocol works correctly
-
40: sparse-mode protocol works correctly
-
20: analysis (write-up)