interactive gaming in wireless environments
TRANSCRIPT
Interactive Gaming in Wireless Environments
Padova, 7 November 2008
Claudio E. [email protected]
Dipartimento di Matematica Pura e Applicata Università degli Studi di Padova
Outline
• Introduction
• Fast and Fair synchronization (FILA)
• Wireless last hop scenario
• Gaming in VANETs
• Conclusion
Multiplayer Online Games (MOGs): Why?
• Very challenging research area
• Shares a family of problems with several traditional research fields in computer science:– Distributed simulations– Military simulations– Virtual reality– Safety ensuring applications– Homeland Security
• Exploding market
Main Problem: Synchronization Offset
GREEN PLAYER’S VIEW BLUE PLAYER’S VIEW
• Armagetron: game exploiting clientserver architecture• Different delays between each player and the server
– The green player is closer to the server than the blue one– Physical distance but also different congestion levels
MOG’s Requirements
• Scalability: – the number of contemporary players should not be bounded– Massively Multiplayer Online Games (MMOGs)
• Interactivity: – multiplayer gaming is extremely delaysensitive
• Consistency:– uniformity of game state view in all nodes
• (Network) Fairness: – simultaneous game evolution regardless of network conditions
Our Approach for Infrastructured Wireless Networks
architecture + synchronization scheme + “Smart” AP
Fairness and Interactivity Loss Avoidance (FILA)
Mirrored GameServer Architecture
1. Fast synchronization among servers
2. Delay equalizationamong players
Avoid queuing delaysdue to concurrent apps
Smart Access Point withLow Advertised Window
(SAPLAW)
Ensure high interactivity and fairness degrees
Geographical/numerical scalability, simplicity,
and centralized control
Core of the system Edge of the system
Internet
Game Constraints Relaxation
• Semantics of the game used to relax constraints (e.g. total order and reliability) and gain interactivity
• The importance of certain game events diminishes when fresher game events make them superseded
• Correlated and interleaved events may preserve the importance of game events
p = 0p = 0
• avgOL ∈ [0, min): no events are discarded• avgOL ∈ [min, max): obsolete events are dropped with a probability p• avgOL ∈ [max, ∞): all obsolete events are discarded
FILA’s Main Component: REDbased Dropping Scheme• Synchronization Algorithm able to drop superseded events• At each Server, for each received event, compute an average
Overall Latency (avgOL) value– avgOL = avgOL + w * (sample – avgOL)
avgOLavgOL
Dropping Dropping ProbabilityProbability
PmaxPmax
11
maxmaxminmin
Phase 0Phase 0 Phase 1Phase 1 Phase 2Phase 2
min)(maxmin)(avgOLPmax
p−
−×=p = 1p = 1
Trading Reliability for Interactivity
• Contribution: in case of lousy interactivity then superseded game events may be dropped to speed up processing
• Consistency of the game state is fully maintained
0
0.2
0.4
0.6
0.8
1
1.2
0 30 60 90 120
150
180
210
240
270
300
330
msec
Prob
(GTD
< x
)
OFF ONOFF ILA
SynchOffset cumulative function
Prob
( Sy
nch
Off
set <
x)
Also real experiments
Wireless Scenario: Last Hop• Homes are common places where
enjoying different entertainment services– TiVo, Media Center (Home Entertainment
Center)
• Vehicles will soon be generally endowed with entertainment hubs
• Increasing available wireless bandwidth– E.g., IEEE 802.11g, IEEE 802.11p
Problem: Delays on the Last Hop• Wireless is prone to errors, retransm., disconnections
• Wireless is a naturally shared environment– Many, heterogeneous flows may compete for the same channel
• General Belief: UDP’s lack of congestion control harms TCP flows
• Problem & Contribution: we showed that even TCP harms UDP flows– Window based flow control mechanism– Continuously probe the link for more bandwidth– Can fill up links and buffers with its packets– Increases delays cwnd
time
pipe capacity
Problem Statement: FTP Impact on Concurrent RealTime Applications
time (sec)
jitte
r (m
sec)
VIDEO STREAM
ONLINE GAME
Starting times
VIDEO CHAT
FTP/TCP DOWNLOAD
Concurrent TCP traffic jeopardizes interactivity for online
applications
Smart AP with Limited Advertised Window: SAPLAW
)(#))((
)(tTCPflows
tUDPtrafficCtmaxTCPrate
−=
Contribution: SAPLAW tries to avoid buffer utilization exploits information available at the AP onthefly modification of the advertised window with appropriate values:
time
cwnd
pipe size
limited cwnd
regular cwnd
time
cwnd
pipe size
limited cwnd
regular cwnd
No queues: lower delays smoother throughput (no losses and reductions of the sending window for TCP flows)
SAPLAW Applied to Vehicular Scenario
Urban scenario – 50km/h – grid topology
Highway scenario – 120km/h – linear topology
Heterogeneous traffic (FTPTCP, online gamingUDP)
1000m between APs
About 750m tx range
IEEE 802.11p
Both static and mobile (vehicle) nodes
Congestion Window Comparison
With SAPLAW no congestion losses
advertised window dynamically changes with the number of TCP sessions on a certain AP
TCP regular
SAPLAW
Jitter Comparison
SAPLAW drastically reduces delays
(while maintaining the goodput)
In Summary: Infrastructured Scenario
• System composed by – hybrid architecture– synchronization scheme– smart AP
• Holistic approach to provide interactivity, fairness, consistency, and scalability
No Infrastructure: Online Gaming in a Car Platoon
• Scenario: online games played by passengers during long trips on a highway
• Absence of infrastructure: VANET (Vehicular Adhoc Network)
• Game sessions of few minutes or tens of seconds (e.g., Armagetron)
AB
D
H IJ
C
F
GK
E
Assume that A sends a game message to all other cars
Gaming on VANETs: Problem Statement
• Game Messages have to be delivered very quickly to all cars in the gaming car platoon.
• Problems arise from multiple transmissions– multiple forwarders
• Various proposals to reduce multiple (and redundant) transmissions.
Example of Optimal Solution
• Who has to forward a certain game message?– Optimal solution: A, C, F, G, and F broadcast the
game message (one message and 4 hops)
AB
D
H IJ
C
F
GK
E
Assume that A sends a game message to all other cars
Related Works
• Vehicles have priorities in becoming the next forwarder – inverse proportion of the distance from the sender
• Problem: limitation of preceding works:– Require perfect knowledge of the topology– Waste time in determining which is the last vehicle in the
transmission range.– Unrealistically assume that there is a unique, constant, and
known a priori transmission range for all cars in every moment• Problem: in a realistic environment they would use wrong
parameters, thus not behaving efficiently (delays, losses).
Fast MultiBroadcast Protocol: Basics
• We designed Fast MultiBroadcast Protocol (FMBP) to have Game Messages covering the areaofinterest in as less time as possible (as few hops as possible)
• Contributions of FMBP:– Each vehicle dynamically computes its own transmission range– Tx range estimation is put to good use to reduce the number of
hops that a Game Message will experience in its trip to destination.
FMBP: the Estimator• In Game Messages:
– Position– Tx range– Hearing distance
3. Game Message including B’s range estimation
2. Game Messages also say how far C can hear
1. Through being able to hear A’s Game Messages, C knows its “hearing distance”
A B C
• Collected info are periodically refreshed
Contention Window Computation• Cars broadcast Game Messages containing their Estimated Transmission Range and position
• Cars forward the Game Message after a contention window inversely proportional to the distance from the source:
• If another car that is farther from the source than the considered one already forwarded the Message, then the considered car aborts its sending procedure (the message has already been propagated)
+
−×−
CWMin)CWMinCWMax(MaxRange
cetanDisMaxRange
Simulation Assessment
• Stripshaped road (highway)• Areaofinterest of 8km. • Within the same simulation, cars have speeds
uniformly distributed in the range 72144Km/h. • CWMin and CWMax equal to 32 and 1024 slots
– as the real IEEE 802.11 protocol• Two possible cases for the actual transmission
range: 280m and 1000m.
Compared Schemes
• Fast Broadcasting– Exploits our tx range estimator
• Static300– considers 300m as a fixed parameter for the tx range– Ideal iff the actual transmission range is indeed 300m
• Static1000– considers 1000m as a fixed parameter for the tx range– Ideal iff the actual transmission range is indeed 1000m
Summarizing Results: 280m of Actual Transmission Range
• FMBP ensures a 200ms of average delay to deliver a game event from one extreme to the other of the gaming car platoon
• Static schemes are outperformed
FMBP is less sensitive to cars’ dispersion in the VANET
(now the last car in the tx range is in average at 240m instead of 280m)
Summarizing Results:Other Configurations
FMBP has the same performance than the ideal scheme (Static1000) on a configuration with 1000m of
transmission range
Conclusion
• To support online gaming even for wireless and highly mobile users we proposed and evaluated:– synchronization mechanism based on an Internet
server overlay that enables scalability, fairness, and interactivity;
– Smart AP to guarantee elastic (TCP) and realtime (games) coexistence on a wireless link;
– a fast multihop broadcast scheme that exploits a novel transmission range estimator to efficiently deliver game events to players in vehicles.
Other Works
• Transport protocols design – (e.g., satellite links, RTTfairness)
• Network parameters estimators– Capacity, bandwidth, shareable bandwidth
• Vehicular networks– MAC Protocols, Transport Protocols, applications– Infrastructured, ad hoc
• Multimedia over wireless networks– Gaming, videostreaming– Even on Mesh networks
• Altruisticaimed web applications• Computerbased preservation of Cultural Heritage