Jose Saldana, Julian Fernandez Navajas, Jose Ruiz Mas, Luis Sequeira, Luis Casadesus, "Comparison of Multiplexing Policies for FPS Games in terms of Subjective Quality". Proc. II Workshop on Multimedia Data Coding and Transmission 2012, Jornadas Sarteco. Elche (Spain). Sept. 2012. ISBN: 978-84-695-4472-3
Comparison of Multiplexing Policies for FPS Games in terms of Subjective Quality
1. COMPARISON OF
MULTIPLEXING POLICIES FOR
FPS GAMES IN TERMS OF
SUBJECTIVE QUALITY
GTC
Communication
Technologies Group
Jose Saldana
Julián Fernández-Navajas
José Ruiz-Mas
Luis Sequeira
Luis Casadesus
University of Zaragoza, Spain
9. Introduction
Scenarios where a number of players
share the same connection
TCM
TCM
Multiplexer
.
.
.
TCM
Multiplexer
Central
Server
Players
Access
Multiplexer router
Game
Server
10. Introduction
By multiplexing, we can save
- Bandwidth
- Packets per second
delaymux
delayrouter
delaynetwork
IP network
router
.
.
.
MUX
DEMUX
Game Server
Players
IP
TCM
IP
11. Introduction
Adaptation of an RTP VoIP mux
technique, to non-RTP flows
payload
payload
ECRTP
...
ECRTP
RTP
UDP
IP
PPP Mux
PPP
L2TP
VoIP
IP
One IPv4/UDP/RTP VoIP packet with two samples of 10 bytes
η=20/60=33%
40 to 6-8 bytes compression
Five IPv4/UDP/RTP VoIP packets with two samples of 10 bytes
η=20/60=33%
One IPv4 TCMTF Packet multiplexing five two sample packets
η=100/161=62%
saving
12. Introduction
Adaptation of an RTP VoIP mux
technique, to non-RTP flows
One IPv4/TCP packet 1500 bytes
η=1460/1500=97%
Payload
Payload
Reduced Header
...
UDP
Reduced Header
IP
PPP Mux
PPP
One IPv4/UDP/RTP packet of VoIP with two samples of 10 bytes L2TP
η=20/60=33%
IP
One IPv4/UDP server-to-client packet of Counter Strike with 9 players
η=160/188=85%
Four IPv4/UDP client-to-server packets of Counter Strike
η=61/89=68%
One IPv4/TCM packet multiplexing four client-to-server Counter Strike packets
η=244/293=83%
saving
14. Introduction
By multiplexing, we can save
- Bandwidth
- Packets per second
… at the cost of adding
- Delay
- Jitter
delaymux
delayrouter
delaynetwork
IP network
router
.
.
.
MUX
DEMUX
Game Server
Players
IP
TCM
IP
15. Introduction
Two policies to define which packets
are multiplexed
1) period
Native
traffic
...
PE
PE
PE
Multiplexed
traffic . . .
2) timeout
Native
traffic
...
Multiplexed
traffic . . .
PE
...
...
TO
TO
...
...
16. Introduction
Expected results:
Period
- Smaller savings
- Less jitter
Native
traffic
...
PE
PE
PE
PE
...
Multiplexed
traffic . . .
...
TO
Timeout
- Higher savings
- Higher additional jitter
Native
traffic
...
Multiplexed
traffic . . .
TO
...
...
17. Introduction
In this work, we compare timeout and
period policies, in terms of a subjective
quality estimator.
Tradeoff: savings vs jitter
19. Test methodology
- Traffic of the game
-
Small packets (79.5 bytes avg)
64 pps
40 50 60 70 80 90 100 110
bytes
0
10 20 30 40 50 60 70
ms
20. Test methodology
- Simulation scenario:
-
-
Traces of gaming traffic
Background traffic
RTT delay: sum of the delays
delaymux
delayrouter
delaynetwork
IP network
router
.
.
.
MUX
DEMUX
Game Server
Players
IP
TCM
IP
22. Test methodology
BW↓
PE/TO ↑
Buffer
delay and
jitter↓
Mux delay
and
jitter↑
delaymux
delayrouter
delaynetwork
IP network
router
.
.
.
MUX
DEMUX
Game Server
Players
IP
TCM
IP
23. Test methodology
- E-Model: VoIP delay and packet loss
- FPS games: different studies
consider delay limits, and also
packet loss limits
- G-Model: MOS formula for Quake
IV, adapted from E-Model: delay
and jitter. Packet loss is not
considered under 35%.
25. Tests and Results
5 players TO
30
5 players PE
25
Average Retention Time
10 players TO
5 and 10 players: TO
adds more delay
10 players PE
ms
20
15
10
5
0
5
10
15
20
25
30
35
Period or timeout (ms)
40
45
50
26. Tests and Results
5 players TO
30
5 players PE
25
Average Retention Time
10 players TO
10 players PE
ms
20
15
10
Saturation: above 25 ms,
a size of 1500 bytes is
reached, so the packet is
sent
5
0
5
10
15
20
25
30
35
Period or timeout (ms)
40
45
50
27. Tests and Results
15 players TO
30
15 players PE
25
Average Retention Time
20 players TO
15 and 20 players: slight
difference
20 players PE
ms
20
15
Retention: T/2
10
5
0
5
10
15
20
25
30
35
Period or timeout (ms)
40
45
50
28. Tests and Results
Retention time histogram TO=15ms
number of packets
1000
800
600
400
200
0
0
5
10
Timeout (ms)
15
20
15
20
Retention time histogram PE=15ms
number of packets
1000
800
600
400
200
0
0
5
10
Period (ms)
29. Tests and Results
Peak of 4119
packets: trigger
Retention time histogram TO=15ms
Tail above 15 ms: more
jitter. No upper bound
for delay
number of packets
1000
800
600
400
200
0
0
5
10
Timeout (ms)
15
20
15
20
Retention time histogram PE=15ms
number of packets
1000
800
600
400
200
0
0
5
10
Period (ms)
30. Tests and Results
20 players PE
Retention Time stdev
15 players PE
18
10 players PE
5 players PE
16
20 players TO
15 players TO
14
10 players TO
5 players TO
12
stdev (ms)
10
8
6
4
2
0
5
10
15
20
25
30
Period or Timeout (ms)
35
40
45
50
31. Tests and Results
20 players PE
Retention Time stdev
15 players PE
18
10 players PE
5 players PE
16
20 players TO
15 players TO
14
10 players TO
5 and 10
players: higher
difference
5 players TO
12
stdev (ms)
10
8
6
15 and 20 players: slight
difference
4
2
0
5
10
15
20
25
30
Period or Timeout (ms)
35
40
45
50
32. Tests and Results
Next tests:
- Background traffic
- 5 players
- 5, 15, 25 ms period / timeout
38. Tests and Results
MOS, Quake IV, 5 players, 10kB
4
3.5
MOS
3
2.5
2
PE is globally better. Smaller retention
time and jitter are better than higher
bandwidth saving
PE=5ms 10kB
TO=5ms 10kB
PE=15ms 10kB
TO=15ms 10kB
PE=25ms 10kB
TO=25ms 10kB
1.5
1
0
200
400
600
800
1000
1200
1400
background traffic (kbps)
1600
1800
2000
39. Tests and Results
The difference can only be appreciated
when the number of players is small
Number of players
MOSperiod
MOStimeout
Difference (%)
5
3.43
3.32
3.31 %
10
3.37
3.34
0.98 %
15
3.30
3.28
0.42 %
20
3.19
3.19
0.10 %
41. Conclusions
- Two multiplexing policies have
been compared
- This comparison has to be done
in terms of subjective quality,
integrating all network
parameters
- Timeout saves more bandwidth
- Period adds less delay and jitter
42. Conclusions
- MOS shows a slight advantage
for period policy
- The difference can only be
appreciated when the number of
players is small
- The decision may also be
influenced by implementation