Video Streaming over DiffServ Network

advertisement
Video Streaming over DiffServ
and some other Issues
Presented by Wei Wei
Outline




A framework for video streaming over
DiffServ network
A novel node mechanism for video
streaming over DiffServ network
A case study for aggregation and
conformance in DiffServ network
Pricing, Provisioning and Peering of
DiffServ network
Framework for Video over DiffServ[1]

This paper presents a framework for qualityof-service mapping between categorized
packet video and relative DiffServ network.



RPI-based video categorization
QoS mapping under a given cost constraint from
video categories to DS classes
Adaptive packet forwarding mechanism
RPI-based video categorization

Different video factors are taken into
consideration for the RLI association of
video packet.



magnitude and direction of the motion
vector for each MB
encoding types (intra, intra-refreshed, inter,
etc)
initial error due to packet loss
RPI-based video categorization


Combine different video factors by
different weight to get RLI of a packet
RLIs are categorized into K DS
categories using Nonuniform
quantization of RLI
QoS mapping from video
categories to DS classes

Formulated into the following
optimization problem
K 1
min
qk
 QD
k 0
K 1
k
 min ( RLI k  lq (k )  nk )
qk
K 1
p
k 0
q
k 0
(k )  nk  P
QD is quality degradation, RLI k average loss
effect
QoS mapping

Equals to minimize the Lagrangian
formula
J k ( )  [ RLI k  l q ( k )    p q ( k ) ]  nk
QoS mapping
Adaptive packet forwarding
mechanism
Adaptive packet forwarding
mechanism


Combination of adaptive WFQ scheduling
and RED with in/out bit (RIO)
Adaptive WFQ

Providing persistent service differentiation
 i q j (t )

wi (t )  j qi (t )
w j (t )
Conclusions (1)


RPI plays a bridging role in enabling the
network to be content-aware
Adaptive packet forwarding mechanism
provides more persistent network DS levels
regardless of network load fluctuation
A novel node mechanism for video
streaming over DiffServ[2]


Define two types of services: High Reliable (HR)
service, Less Assured (LA) service
Propose a node mechanism called Selective
Pushout with Random Early Detection (SPRED)
A novel node mechanism

Design objectives:




Core router does not maintain per-flow state
Packet sequence within each flow should not be
altered at a node
Packets from HR service should be delivered as
reliably as possible
Packets from TCP traffic should be dropped
randomly during congestion to avoid global
synchronization
A novel node mechanism


Combination of RED with In/Out and
Pushout (PO)
Architecture
A case study for aggregation and
conformance in DiffServ[3]


Motivation: To better understand the impact
of traffic aggregation on conformance.
The focus was on identifying the level of
non-conformance that is introduced into
initially conformant streams after crossing a
DiffServ domain.
A case study for aggregation and
conformance in DiffServ[3]

Scenario
A case study for aggregation and
conformance in DiffServ

Results and Conclusions:

Basic configuration

simply decreases network load does not appear to be
very effective at ensuring egress conformance


Neither the number of cross-traffic streams nor the
number of network hops traversed by the tagged
stream appear to have a major influence on egress
conformance
Two approaches to absorb perturbations introduced
by network interferences: reshaping buffer and
increase of egress token rate. Increasing egress
token rate is much less effective than buffering.

Variability in Packet Sizes


In the presence of variable size packets, an initially
conformed stream of packets can be deemed nonconformant on egress, even without any network
interferences
The dominant effect in terms of the egress
conformance of a stream is its internal packet
variability. In other words, variations of packet sizes
within a stream have a more pronounced effect than
the potentially larger network perturbations caused
by variable packet sizes in cross streams

Aggregate Contracts


The number of streams being aggregated, the
number of cross streams, the number of network
hops being crossed, and the load on network links,
all interact with each other in determining the egress
conformance
Trade-off that exists between the greater level of
network perturbations that higher hop counts or
number of cross streams induces, and the greater
likelihood that aggregate bursts will be broken-up as
they traverse the network

From simulation, we can see that, increasing in hop
count typically improve performance, while
increasing in the number of cross streams usually
degrade performance
Pricing, Provision and Peering


This paper presents a decentralized auctionbased approach to pricing of edge-allocated
bandwidth in a differentiated services
Internet.
In the framework, they have one rawcapacity seller per network, one broker per
service, and users, to act as whole-sellers,
retailer and end-buyers.
Pricing, Provision and Peering

dynamic market-pricing
bandwidth


of
edge-allocated
They use game theoretic analysis to get the optimal
strategies for buyers and brokers.
the feasibility of maintaining consistent service
level agreements across interconnected networks


The good news is that dynamic market-driven
partitioning of network capacity among services
appears sustainable.
The bad news is that very conservatively provisioned
services can be unstable
References



“Quality-of-Service Mapping Mechanism for Packet
Video in Differentiated Services Network,” J. Shin, J.
Kim, and C.-C.J. Kuo, IEEE Transactions on
Multimedia, Vol. 3, No. 2, June 2001
“A differentiated services architecture for multimedia
streaming in next generation Internet,” Y.T. Hou, D.
Wu, B. Li, T. Hamada, I. Ahmad, and H.J. Chao,
Computer Networks 32, 2000, pp 185-209
“Aggregation and Conformance in Differentiated
Service Networks: A Case Study,” R.A. Guerin and V.
Pla
Download