1. Armin Bahramshahry August 2009

2.  Background  Problem  Solution  Evaluation  Summary

3. virtual world avatars guns swords Simulate a virtual world, and represent players as avatars with cool guns and swords

4. Internet multiple converge  Run multiple game instances and desire to converge to the same game state TypeCommunicated DataAllow Discrepancy DeterministicPlayer action  Non-deterministic Changes in game object properties

5.  Traditional Challenges > game traffic ◦ Limited upload capacity & communication delay  New Challenges > heterogeneous traffic Internet Player Action Event QueuesMessage Queues Event Feature Latency Sensitive Require Reliability Require Order VoIP File transfer Text chat

6. Game Updates Game Network Services Predictive Contract Mechanism (PCM) Area-of-Interest (AOI) Network Quality of Service (e.g., OverQoS)

7. Game Update VoIP Traffic File Transfer Network Services Game VoIPFile Optimizations Network Services Services & Information ? Requirements MCCA Game  Simplify development  Key Idea #1: Divide and Conquer ◦ Efficient use of network resources  Key Idea #2: Cooperate with the Game ◦ Support for new game level optimizations ◦ Enable game adaptation to network conditions Services

8. KEY #1: Divide and Conquer  Differentiated Services  Class Definition ◦ Priority ◦ Order & reliability ◦ Loss rate ◦ Latency requirements ◦ Bit-rate throttling GameVoIPFile Optimizations Network Services Differentiated Services ? Requirements MCCA Game

9. KEY #2: Cooperate with the Game  Network Condition Reports (local) ◦ Observed network conditions: Loss, latency, jitter  Group Communication Policy  Quality Estimation  Distributed Quality Aggregation (global) GameVoIPFile Optimizations Network Services Differentiated Services Message Forwarding Reports Quality Requirements Quality Estimation MCCA Game ? Requirements & Quality

10.  Perceived QoS is Class Dependent  Game ◦ Estimate game update quality ◦ Based on observed loss and delay  VoIP ◦ Based on télécommunication standard (E-model)  File Transfer ◦ Transfer rate Game Message Quality Latency Max Quality No Quality (α) latency - 100ms for Quake III (β) latency – 200ms for Quake III GameVoIPFile Quality Estimation Optimizations Quality Network Services Differentiated Services Message Forwarding Reports Requirements & Quality MCCA Game

11.  Optimistic Game ◦ Uses negative acknowledgments (i.e., N-consecutive message losses) ◦ Avoids large updates and wasteful acknowledgments  Adaptive Game Update Interval ◦ Dynamically sets the game update interval (based on the estimated observed quality) ◦ Reduces network traffic due to game updates  VoIP Optimization ◦ Disables VoIP at low qualities ◦ Avoids useless VoIP traffic GameVoIPFile Quality Estimation Optimizations Quality Network Services Differentiated Services Message Forwarding Reports Requirements & Quality MCCA Game

12. VoIPFile Quality Estimation Optimizations Quality Network Services Differentiated Services Message Forwarding Reports Requirements & Quality MCCA Game Real Network Network Emulator Network Simulator

13.  Game Update Generator ◦ First person shooting game, Quake III (based on the model by Bharambe et al., SIGCOMM’08) ◦ Two-frame delta encoding ◦ 20 game updates per second  VoIP Traffic ◦ On/Off Markov process (based on the model by Padungkrit et al.) ◦ 50 VoIP messages per second  File Transfer ◦ At uniformly distributed intervals GameVoIPFile Quality Estimation Optimizations Quality Network Simulator Differentiated Services Message Forwarding Reports Requirements & Quality MCCA Game

14.  Operates at the message level  Provides point-to-point communication  Does not simulate cross network traffic ◦ Upload capacity << Network core (Bharambe et al.)  Network Simulator Validation: ◦ Compare to a packet-level network emulator GameVoIPFile Quality Estimation Optimizations Quality Network Simulator Differentiated Services Message Forwarding Reports Requirements & Quality MCCA Game

15.  Workload Combinations  Game Level Optimizations  Class Definitions ◦ Window-size ◦ Redundancy level and redundancy interval ◦ Forwarding policy  Workload Definitions  Real-Time Changes in: ◦ Number of players ◦ Upload capacity  Deployment Environment  …

16. New game-level optimizations!

17. Efficient use of network resources & new game-level optimization!

18. Game can adapt!

19.  Better gaming experience using: ◦ Prioritization ◦ Classification ◦ Targeted QoS ◦ Efficient group communication  Distributed quality aggregation enables game adaptation to network conditions  Lower quality variability among players

20.  Demonstrated the benefits of: ◦ Differentiated services ◦ Cooperation with the game  Prototyped a generic communication layer for online games (MCCA) ◦ Experimental evaluation of MCCA at the communication layer to improve player’s experience.

21.  Outcome: ◦ Paper submission to NetGames 2009 ◦ Network Engineer @ United Front Games Other Publications:  “vanDisk: An Exploration in Peer-To-Peer Collaborative Back-up Storage”, Amir Javidan, Tony Angerilli, Armin Bahramshahry, Guy Lemieux, Roman Lisagor, Matei Ripeanu, 20th IEEE Canadian Conference on Electrical and Computer Engineering, Vancouver, BC, April 2007.  “A High-Performance GridFTP Server at Desktop Cost”, Samer Al-Kiswany, Armin Bahramshahry, Hesam Ghasemi, Matei Ripeanu, Sudharshan S. Vazhkudai, poster, SC2007, Reno, NV, November 2007. GameVoIPFile Optimizations Network Services Differentiated Services Message Forwarding Reports Quality Requirements & Quality Quality Estimation MCCA Game