SPANNING TREE CONSTRUCTION AND HOP-COUNT
========================================

All the sensor network projects used one of the following two
approaches, both of which are reasonable.

(1) Periodic flooding: Periodically, the root node would initiate a
flooding of a message, which would be further propagated by the other
nodes, together with a hop-count.  When a node first receives this
message, it will set the sender of the message to be its parent,
increment its hop-count and then forward the message out.

Sequence number: The root adds a new sequence number for every
flooding it initiates.  When a node receives a message, it updates its
parent info if the sequence number is newer or if the sequence number
is the same as the most recent one it received and the hop-count
received is smaller than the best it has received thus far.

Timeout: The hop-count should time out so that a node does not keep
stale information.  It is best to set the timeout to be greater than
the period of the flooding procedure.

(2) Distance vector: This is the classic distance vector protocol in
which nodes periodically broadcast their distance (from the root) to
their neighbors.  And on receipt of such a message, a node updates it
parent and/or hop-count if the new hop-count is smaller than the
current one it has.

Sequence number: One could have the root node initiate the distance
vector procedure periodically with a new message using a new sequence
number.  This would make the procedure quite similar to periodical
flooding.

Count to infinity: As we all know, one problem with the basic distance
vector approach is the count to infinity problem, when a subset of
nodes gets disconnected from the root.

Timeout: Using sequence numbers and having each node periodically
timeout its entry (even in the presence of updates) would address the
count to infinity problem.

DYNAMICS
========

Both of the above approaches work well in the dynamic setting if
timeouts and sequence numbers are appropriately set.

ALARM
=====

Given the spanning tree setup, the alarm implementation is to simply
propagate the alarm to the root using the path through the parent.  A
couple of issues need to be kept in mind:

Periodic propagation of alarm: Since communication is unreliable, it
is important that the alarm source continually send the alarm.  This
period should be much smaller than the period used in the spanning
tree construction so that alarm is relatively stable.

Culmination of alarm: Some of you implemented a "stop alarm" message.
In the absence of reliable delivery (through ACKs, etc.), this is not
a good idea since a loss of a single message kills this step.  When
the periodic propagation of alarm stops, then this implicitly signals
the end of the alarm.

COLLISION AVOIDANCE
===================

It is a good idea to use random backoff periods for avoiding
collisions.  For networks of small size, there is no need to implement
RTS/CTS or ACKs.

TIMING
======

Appropriate selection of timers and timeout values turned out to be
quite critical for the project.  For instance, it makes sense to set
the period of DV or flood protocol to be more than the alarm
generation period.  Also, one should avoid making one timer a multiple
of the other (or share a large common divisor) since handling events
that may simultaneous fire is unpredictable.  Use of timeouts and/or
sequence numbers to make sure that the network does not maintain stale
information and responds well to dynamics.

GRADING
=======

Allotted points:

3 for explanation/documentation (Exp)
3 for static shortest paths (SSP)
2 for dynamics (Dynamic)
2 for alarm (Alarm)

Your score:
				Exp	SSP	Dynamic	Alarm

Gil+Kaushik+Mike		3	3	2	2
Zhifeng+Abhishek+Jingjing	3	2.5	2	1.5
Emrah+Ana-Maria+Maulin		3	3	2	1.5
Vrushali+Priyanka+Sudha		2.5	3	1.5	1
Gokul+Vishal+Chirayu		2	2.5	1	1
Anand+Tejas			3	3	1.5	1.5
Bishal+Jia+San			3	3	2	2
Utkarsh				1.5	1	0	0