Transcript Differentiated Services

Extending an Open MediaStreaming Platform to Support
Differentiated Services
S.Zoi D. LouKatos, P. Papageorgiou, P.
Stathopoulos, N.Mitrou
Ece Department
National Technical University of Athens
Greece
1
Outline






Introduction
Related Work
DiffServ Aware Architectural Extensions
Quality Evaluation Metrics
Experimental Evaluation of the Enhanced Platform
on a DiffServ Lab Testbed
Future Work
2
Introduction



State-of-the-art multimedia technology gives the
potential to author complex networked multimedia
applications, composed of multiple media streams.
Although network level QoS do exist, the deployment
of QoS mechanisms to such applications is still very
limited.
This is mainly because the definition of portable QoS
specifications is still an open issue.
3
Introduction (contd.)


There are several application level QoS frameworks
have appeared so far (e.g. DMIF and Winsock 2 API),
but studies are restricted to single stream traces, and
simulated networks.
What is needed is the ability of the application to
define its own quality parameters, in such a way that
perceived quality is also captured.
4
Introduction (contd.)

QoS Services
 Best-Effort Best-Effort does not provide QoS,
because there is no reordering of
packets.
 Differentiated Services
(DiffServ) DiffServ, as the name suggests,
differentiates between multiple traffic
flows.
 Integrated Services
(IntServ) IntServ is often referred to as “Hard
QoS,” because it can make strict
bandwidth reservations. Needs
signaling first. Must be configured on
every router along a path.
5
Introduction (contd.)

In order to support the quality of differentiated
service in the IP network, DS Field (DiffServ Field)
was proposed by IETF in RFC2427.
DSCP : DiffServ Codepoint
CU : Currently Unused
6
Introduction (contd.)


The DiffServ network uses DSCP to classify data
packets and forwarding them according to the
corresponding PHB (Per Hop Behavior).
IETF has proposed three kinds of PHB :
 DF (Default Forwarding) PHB
- also called Best-Effort , DSCP : 000000
 EF (Expedited Forwarding) PHB
- also called Premium service , DSCP : 101110
 AF (Assured Forwarding) PHB
- AFij , 1<=i<=4 : class
, 1<=j<=3 : drop precedence
7
Related Work


The differentiated Services (DiffServ) framework
supports the differentiation of packets, not only
belonging to different competitive streams, but also
within the same stream.
This is achieved by marking the ToS byte of each
packet header.
8
Related Work (contd.)

In MPEG-4 encoding, high compression is achieved at
the cost of low error resistance and several error
resilience mechanisms are foreseen by the standard
such as resynchronization and data partitioning.
 resynchronization
 data partitioning
9
Related Work (contd.)



At user level, perceived quality is evaluated by using
subjective methods. However, these can be costly,
and time-consuming to give quantifiable results.
In the case of audio a set of methods of quality
evaluation have been proposed, and they are take
into human psychoacoustics.
In the case of video, many evaluation of the pictures
based on the Peak Signal-to-Noise Ratio (PSNR).
10
Related Work (contd.)

MPEG4IP
 The MPEG4IP project is an open-source platform,
incorporation additional open source tools from other parties.
 The client side mainly comprises the player and the decoders,
while the server side comprises the following components :
- A toolkit for off-line encoding of MPEG-4 compatible streams.
- An application (mp4live) for capturing, real-time encoding
and streaming video and audio content.
- A streaming server, the open source Apple`s Darwin
Streaming Server (DSS)
11
DiffServ Aware Architectural
Extensions

The QoS Framework for
real application -“QCompiler”
is adopted in this work.
12
DiffServ Aware Architectural
Extensions (contd.)

High-level application specification layer
 user quality are defined, for example High, Medium, and Low.
 With a quality study component has been developed inside the
MPEGIP platform, QoS violation effects (e.g. losses, delays) can
be simulated and previewed on a media stream before this is
transmitted to the network.
13
DiffServ Aware Architectural
Extensions (contd.)

Next, the quality degradation observed in a preencoded video stream, in the following scenarios :
 Scenario 1 (no error resilience) : The 5th packet of the first I
frame is lost.
 Scenario 2 (no error resilience) : The last packet of the first I
frame is lost.
 Scenario 3 (with error resilience) : The 5th packet of the first I
frame is lost.
 Scenario 4 (with error resilience) : The last packet of the first I
frame is lost.
 Scenario 5 (with error resilience) : The last packet of the 3rd P
frame is lost.
14
DiffServ Aware Architectural
Extensions (contd.)
15
DiffServ Aware Architectural
Extensions (contd.)

The Metadata Compilation Layer and the Binding
Layer
 In order to enhance mp4live application with packet
differentiation extensions, a dedicated marking component was
implemented.
 This component is capable of differently labelling parts of the
produced bitstream with arithmetic values corresponding to
different quality levels.
 This values will be assigned to the corresponding packets as
ToS values by the binding layer, during packetization.
16
DiffServ Aware Architectural
Extensions (contd.)
VOP : Video Object Plane
17
DiffServ Aware Architectural
Extensions (contd.)

The Run-Time Metadata execution layer
 metadata descriptions are parsed and different QoS levels are
assigned to different network classes, i.e. ToS mapping to
DifferServ classes.
18
Quality Evaluation Metrics


It becomes apparent that packet level metrics (e.g.
% packet loss rate) cannot give an accurate
estimation of the final perceived quality.
In the case of video, spatial and temporal effects of
losses on frames are of interest to map user
perceived quality.
19
Quality Evaluation Metrics
(contd.)



In the case of Internet telephony, the perceived
quality after bursty packet losses is often worse than
when the occurrences of these losses are adequately
spaced.
To address this problem, the finite horizon QoS
criterion call (m,k)-firm guarantee was proposed.
In the case of video streams, the (m,k) criterion is
meaningless at packet level. On the contrary it is
meaningful at frame level, as the frame is the entity
mostly perceived by human eye.
20
Experimental Evaluation of the
Enhanced Platform on a DiffServ Lab
Testbed





Experiments Platform topology
MPEG4IP QoS extention
DSS
a DiffServ enabled router (AF PHB implemented)
MPEG4IP QoS extention
21
Experimental Evaluation of the
Enhanced Platform on a DiffServ Lab
Testbed (contd.)
 Encoding parameters of a video stream :
Frame rate :10fps, Resolution : 176 X 144, Encoding rate : 500kbps
22
Experimental Evaluation of the
Enhanced Platform on a DiffServ Lab
Testbed (contd.)


The video stream encoded with and without error
resilience, is experimentally evaluated under Best
Effort and DiffServ configurations.
Error resilience mechanisms are based on
resynchronization markers, for I frames and data
partitioning for P frames.
23
Experimental Evaluation of the
Enhanced Platform on a DiffServ Lab
Testbed (contd.)

Queue statistic
24
Conclusions and Future Work


More sophisticated packet-marking policies for
unequal protection based on stream semantics, such
as objects extracted with video segmentation
algorithms will be explored.
The use of metadata descriptions for defining
portable QoS specification will also be explored within
this QoS framework based on content description
technologies, such as XML.
25