Transcript Project Presentation

Transmitting Scalable Video over
a DiffServ network
EE368C
Project Presentation
Sangeun Han, Athina Markopoulou
3/6/01
Sangeun Han, Athina Markopoulou
1
Project Proposal
• Problem:
– Video transmission over the heterogeneous Internet
• Facts:
– Scalability: different parts of a video stream contribute
unequally to the quality.
– DiffServ Networks can provide service differentiation,
based on the marking of packets.
• Proposal
– Limit the effect of loss when it happens. Prioritize
information according to importance and drop packets
accordingly.
Sangeun Han, Athina Markopoulou
2
Specifics
• What type of scalability? H.263+, SNR
EL
EI
EP
EP
EP
BL
I
P
P
P
• Which DiffServ class? AF (priority dropping)
buffer
packet
scheduling
conditioning management
EF
classification
AF11
high strict priority
AF1
w2
AF2
w3
AF3
w4
AF4
w5
BE
w6
Sangeun Han, Athina Markopoulou
3
Simulation scenario
Main stream: Foreman (10fps)
136Kbps, BL+EL, 2min
10-20 Interfering Streams
BL+EL~=136Kbps
random parts of 6 different streams
H.263+
Encoder
+
Layering
RTP
Packet.
for H.263
(*)
Single AF queue,
2 levels, 100KB
1.5Mbps
Marker
Depackt.
Loss
info
(*) Mode A: at frame level,
Total header= IP(20)+UDP(8)+RTP(12)+H.263(4)=44B
Decoding+
[Error
Conceal.] (**)
Original
Stream
(**) Freezing previous frame
Sangeun Han, Athina Markopoulou
4
Objective of the Project
• Show the benefit from using Priority Dropping for
Scalable Video
– MUX gain
– Graceful Quality Degradation
– Handle short term congestion
• Configuration
– AF queue:
• buffer management, thresholds, other parameters
– Layering parameters
• base layer, temporal dependence
• Recommendation
– To Feedback or to Drop?
Sangeun Han, Athina Markopoulou
5
MUX gain
Layered+PD
Nonlayered
Sangeun Han, Athina Markopoulou
6
Graceful degradation with loss
FGS
NL, no loss
Layered+loss
+ data loss
Non Layered + loss
Sangeun Han, Athina Markopoulou
7
Short Term Congestion
• The source may react to congestion by adapting its
transmission rate...
Rate
Congestion
EL
BL
Reaction
with no
delay D=0
time
R
time
D
D
Reaction
with Delay
D>0
time
Sangeun Han, Athina Markopoulou
8
Reaction time vs.congestion duration
• Simple example:
– 10 streams + 5 more in [55sec,65sec]
– 10 streams react by dropping their EL in [55+D, 65+D]
Sangeun Han, Athina Markopoulou
9
Heavier congestion
• Heavy + non adaptive interfering traffic:
– 10 streams + 10 more in [55sec,65sec]
– 10 streams react by dropping their EL in [55+D, 65+D]
Sangeun Han, Athina Markopoulou
10
Priority dropping vs Feedback
• Feedback
•
•
•
•
is limited by delay
saves network resources
requires coordination
Priority Dropping
– is like reaction in D=0, by appropriate rate decrease
– may handle non adaptive sources
Rate
EL
Congestion
R(t
)
BL
time
Sangeun Han, Athina Markopoulou
11
Configuration of AF queue
BL - low drop precedence
EL - high drop precedence
Drop
prob
1
• Choices:
0
High drop Low drop
L_min
L_max
H_min,max
Buffer occupancy
– Thresholds for the different priorities
– Buffer management: RED or DropTail?
• Observations:
– Not sensitive to choice of thresholds
– RED inappropriate: do not use Avg Qsize, set Lmin=Lmax
– Differentiation: (I) different thresholds (II) Occupancy
Sangeun Han, Athina Markopoulou
12
RED worse than DropTail
For all loads….
and
…for all thresholds
Sangeun Han, Athina Markopoulou
13
Threshold for EL(HP)
• By assigning the buffer
thresholds
– we control the Queue
Occupancy for BL, EL
Threshold_HDP = 56
Threshold_HDP = 16
Sangeun Han, Athina Markopoulou
14
Threshold for EL(LP)
• …this way we distribute
the loss among BL and EL
• ….and thus the quality
• Insensitive to:
• RED, DropTail
• BL choice
• [more sensitive to load]
Sangeun Han, Athina Markopoulou
15
Effect of BL (I): on quality
degradation
QP(BL)=12, 1:1,
(BL=64kbps:EL=74kbps)
QP(BL)=15, 1:2,
(BL=50kbps:EL=86kbps)
QP(BL)=30, 1:4,
(BL=27kbps:EL=110kbps)
Same target rate:
BL+EL~=136kbps
Sangeun Han, Athina Markopoulou
16
Effect of BL (II): on thresholds
QP(BL)=12, 1:1,
(BL=64kbps:EL=74kbps)
QP(BL)=15, 1:2,
(BL=50kbps:EL=86kbps)
QP(BL)=30, 1:4,
(BL=27kbps:EL=110kbps)
Same target rate:
BL+EL~=136kbps
Sangeun Han, Athina Markopoulou
17
Transmission of Scalable Video
• Use feedback + adaptation at the source to match the
transmission rate with the bottleneck bandwidth, to
save network resources along the path
• Use Priority Dropping to handle short term congestion
Quality
Feedback
BL2
BL1
PD
Rate
loss
Sangeun Han, Athina Markopoulou
18
Future work
• Improvements needed
– realistic feedback + adaptation
– >2 layers
– finish FGS
• New experiments needed
– Delay aspect:
–
–
–
–
–
• Loss at the playback buffer
• Entire streams having different delay requirements
Multiple hops
Single wireless hop (802.11 + QoS)
Video + Data
Larger Bandwidths
Other types of scalability: FGS, Temporal, Spatial, DP
Sangeun Han, Athina Markopoulou
19