Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Branch: master
Fetching contributors…

Cannot retrieve contributors at this time

218 lines (154 sloc) 8.101 kB
** CMPT 431 Distributed Systems
** Richard Vaughan 2013
Instructions for coding project
===============================
* Important * As a build-up to the group project, you must complete an
individual "checkpoint" assignment. This must be all your own
work. You are encouraged to discuss approaches and results, but you
must not share code at this stage.
Checkpoint 1: Speed up Universe by data structure improvements
============================================================
Your task is to speed up Universe by breaking up its single O(n^2)
distance comparison computation into multiple sequential O(m^2)
computations, where m < n. The goal is for the sum of the run time of
all the sequential computations to be less than that of the original.
The problem of measuring the distance between all n robots must be
partioned somehow. We discussed in class some ideas about partitioning
space into contiguous sections. Other approaches exist, such as
hierarchical space decomposition such as quad trees and pyramids
(explained in any computer graphics text). Various clustering methods
may also be used to maintain groups of neighboring robots without a
fixed spatial partition.
You are free to choose any method that interests you, however you have
limited time so choose something that you can implement quickly.
The main distance calculation for one robot to fill its array of
sensors is the Univers::Robot::UpdateSensors() method.
Testing
-------
Universe understands these command line switches:
-? : Prints this helpful message.
-c <int> : sets the number of pixels in the robots' sensor
-d : Disables drawing the sensor field of view. Speeds things up a bit.
-f <float> : sets the sensor field of view angle in degrees
-p <int> : set the size of the robot population
-q : disables chatty status output (quiet mode)
-r <float> : sets the sensor field of view range
-s <float> : sets the side length of the (square) world
-u <int> : sets the number of updates to run before quitting
-w <int> : sets the initial size of the window, in pixels
-z <int> : sets the number of milliseconds to sleep between updates
Your final code must support all these flags: don't disable any
functionality.
Test your code by timing the real time it takes to do a fixed number
of update steps, say 1000 (-u 1000) with a between-step sleep time of
zero (-z 0), for sensor view angles of 270 degrees (the default, or -f
270) and 10 degrees (-f 10), for increasing population sizes [10, 100,
500, 1000, ... up to as large as you like] Like this:
$ time ./universe -d -z 0 -f 270 -u 10000 -p 500
real 1m18.766s
user 1m15.687s
sys 0m0.992s
Run this a few times to obtain an average run time: this is your speed
metric.
(We need to run for many simulation steps since the robot behaviour is
very different at t=0 and t=<thousands> when the sensor FOV is small,
and your run speed may be effected by the robot behaviour. Recall that sensor FOV has a large effect on robot behaviour).
Reporting
---------
1) Write a short report with no more than 2 pages of text plus as many
pages of figures as you wish:
a) explaining the approach and outlining the changes you made to
the code.
b) explaining the new runtime complexity of your Universe.
b) stating your performance claims, supported by timing data. Don't
forget to report variance along with mean run times. Is your
modified universe faster than the original? Is this true for all
population sizes? Why, or why not?
3) Commit your report, code and test scripts to the class repository
along with all instructions needed to run them. Do this before the due
date.
Version control repository instructions for the class are available separately.
Getting Help
------------
Follow the procedures on the class web page for getting help with this
assignment. Do not share code with your colleages, but discussing and
comparing stategies is encouraged.
Group Project
=============
* Form a group of exactly five students. Exactly five. For help with
this concept see http://www.youtube.com/watch?v=xOrgLj9lOwk?t=1m22s
except the value in question is five.
* Register your group on the CMPT courses system http://courses.cs.sfu.ca
Task 1: The one million robot challenge
---------------------------------------
Extend the code of universe to support a very large population that
runs fast. Your challenge is to support one million robots running at
10 updates per second of real time in a large, sparse world.
The courses system provides a Subversion (svn) repository for each
group. See requirement 3 below. Start by committing your initial code
there ASAP.
You should design, implement, and test your system and describe it in
a report of 6-10 pages of text *plus* as many pictures and tables as
you need. Test your system to show the range of parameters in which it
works well, but also beyond, to show when it starts to slow down or
fail. Explain what dynamics you see and why they occur.
Very important: Describe in detail the distributed system problems you
are solving and how you solve them, with proper references to the
class reading, textbooks, online resources, etc.
Your report should also describe how the work was distributed among
the team, and the processes you used to get your software engineering
done (coordination and communication strategies, etc).
Your system should provide a visualization that clearly shows the
robot's behaviour and any aspects of your distributed system you wish
to show in the demo. If your system is widely distributed, this part
can be a significant technical challenge.
In the penultimate week of classes we will agree on parameter sets for
performance comparisons, and will run these as a contest in the final
week. There will be informal prizes and the admiration of your peers
for:
(1) The fastest 1M robot system
(2) The largest population that updates at 10Hz, hopefully > 1M
(3) The most beautiful architecture
(4) The most beautiful visualization
Deliverables:
1. code in the SVN repository, checked in and tagged "submitted".
2. a report in PDF, submitted to the assignment on the Courses system.
3. a 15 minute demonstration session in CSIL with the instructor
and/or TA in the last week of classes.
Required:
1. The outcome must be effectively a single simulated world with the
same behaviour as the original reference code.
2. use three or more computers to generate the simulated world.
3. Use the courses.cs.sfu.ca Subversion (svn) service to develop your
code. Your commit logs will be inspected to ensure that everyone is
participating.
Recommended:
1. Use the CSIL computers. Use as many as you wish, but be considerate
of others particularly towards the end of term.
2. Display detailed information in the visualization for testing,
debugging and demonstrating. Well-presented data are very reassuring
to instructors.
3. Report your progress intermediate results in class and on the
mailing list as a challenge to others. Who can be the first to 100K @
10Hz, etc, on the way to 1M @ 10Hz?
4. compete in the contest in the last week of class.
5. To start, choose either the original code or a fast and/or clean
faster implementation from someone in your group.
Allowed:
1. using third party libraries and frameworks, e.g. Redis, 0MQ.
2. multi-threaded code, as long as at least three processes are run on
three different machines, using threads inside processes is fine.
3. any languages you want, but be careful to meet requirement 1.
Forbidden:
1. sharing your code between groups *UNLESS* discussed with the
instructor in advance and given specific permission. It may be OK to
share tools and supporting code, visualization ideas, etc. but check
with RTV first.
And of course, the obligatory:
2. posting your code online publicly.
3. using code specific to this assignment that you found online.
Questions?
----------
Please ask questions on the courses system forum, and help each other
out with answers. The instructor and the TA will answer messages with
a delay, so jump in and help out if you can.
Jump to Line
Something went wrong with that request. Please try again.