Hazel McKendrick Supervised by Henry Fortuna Distributing Virtual Worlds “How can the processing of autonomous characters in a real-time virtual environment benefit from parallelisation over multiple distributed computer systems?”
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Hazel McKendrickSupervised by Henry Fortuna
Distributing Virtual Worlds
“How can the processing of autonomous characters in a real-time virtual environment benefit from parallelisation over multiple distributed computer systems?”
Project Overview. Processing virtual worlds Single server unsuitable Flexibility, scalability, redundancy
Distributed computing
Project Aim. Create a distributed system Divide a virtual world Update characters in it
Consider:• Processing power• Scalability, power and money saving• Flexibility
Simulation Design. World Maps Characters (thousands)
• A* Pathfinding
System Structure. Server & Worker Node architecture TCP-IP Message Passing
Node thread pool
World Division. Top-down vs Bottom-up
Microcells Smallest unit of world Nodes process several Distributed statically or
dynamically Contain and pass work
Distribution. Optimal assignment
• Can be solved with ILP - Complexity too high Two static approaches used Dynamic algorithm created
Distribution Results. Amdahl's Law 95% to 99% parallelised
0 1 2 3 4 5 6 7 8 9 10 110.000
50.000
100.000
150.000
200.000
250.000
300.000
350.000
400.000
450.000
Processing times with varying numbers of nodes
ServerNode 1Node 2Node 3Node 4Node 5Node 6Node 7Node 8Node 9
Number of computers
Proc
essi
ng ti
me
(ms)
0 1 2 3 4 5 60
0.5
1
1.5
2
2.5
3
3.5
4
Average deviation in processing times, over time
Time elapsed (minutes)
Aver
age
devi
atio
n(m
s)
Scaling Hardware. Virtual world load varies greatly Gustafson's Law Scale Hardware Improvedistribution!
(Simulated)
0 20 40 60 80 100 120 1400
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
0
1
2
3
Minimising HardwareEntities Nodes on
Time (s)
Entit
ies
Nod
es
Evaluation. Scaling nodes Dynamic distribution algorithm
• Over time Microcells Reducing hardware
Conclusion. Reduce processing times Balance load Lower power and cooling costs
Further work• Redundancy• Inter-node communications• Scaling factors
Hazel McKendrickSupervised by Henry Fortuna
Distributing Virtual Worlds
“How can the processing of autonomous characters in a real-time virtual environment benefit from parallelisation over multiple distributed computer systems?”
Related Documents
