ON THE DESIGN OF A SCALABLE MULTIMEDIA STREAMING

SYSTEM BASED ON RECEIVER-DRIVEN FLOW AND

CONGESTION AWARENESS

nigo Urteaga, Iraide Unanue, Javier Del Ser

TECNALIA-TELECOM, Parque Tecnologico, Ed. 202, 48170 Zamudio, Spain

Pedro Sanchez, Aitor Rodriguez

Intelligent Transport Systems and Security, IKUSI-Angel Iglesias, S.A., Paseo Miramon 170, 20009 San Sebastian, Spain

Keywords:

Streaming, Scalable multimedia content, Congestion control, Flow control.

Abstract:

In this position paper we present the design of an end-to-end scalable content streaming system that optimizes

the quality of experience of the end-user by allowing each client to retrieve a customized multimedia stream,

based on both network and client states. By taking advantage of multimedia scalability, our proposed receiver-

driven architecture performs a multilayered streaming, where each client is responsible for controlling the

number of multimedia layers it demands from the server. Furthermore, the streaming system proposed herein

implements both congestion and ﬂow control mechanisms, which are also delegated to the receiver. In order

to properly address both network and client states and restrictions, a set of speciﬁc metrics (Buffer State,

Interarrival Jitter and Loss Event Rate) are utilized, which have been speciﬁcally designed to match the

miscellaneous characteristics of heterogeneous networks and end devices. Built upon such metrics, we present

a decision algorithm that jointly performs congestion and ﬂow control, while maximizing inter-session fairness

and end-user quality of experience. The proposed architecture combines different standard protocols while

guaranteeing independence between components of the streaming system.

1 INTRODUCTION

Multimedia streaming has lately gained momentum

within both industry and academia in light of the

forthcoming redeﬁnition of Internet, mainly moti-

vated by a wide variety of Internet-based applications

envisioned to become vastly demanded in the follow-

ing years. As to mention, in multi-point video confer-

encing each of a number of endpoints require person-

alized versions of a given content, whereas in video-

on-demand the features of the multimedia content de-

livered to each client are established based on service

quality and fees, usable bandwidth, etc.

In this context, as opposed to conventional broad-

cast technologies such as terrestrial or cable televi-

sion, IP (Internet Protocol) networks are inherently

heterogeneous in their underlying communication

This work was supported by the Spanish Centro para

el Desarrollo Tecnologico e Industrial (CDTI) through the

TELMAX project (ref. CEN20071036).

means (i.e. usually composed of combinations of

wired and wireless links with distinct associated com-

munication protocols). This heterogeneity gets even

more involved if one notices that the state, trafﬁc

and characteristics of IP networks usually change

dynamically in time. Besides, the rapidly growing

portable device market has introduced a huge variety

of streaming receivers. Based on this threefold ratio-

nale, scalable multimedia content is called to attain

wide acceptance in the near future, as it provides high

adaptability to all the above scenarios. As research on

scalable content advances with the pioneering Scal-

able (H.264/SVC) and the Multiview (H.264/MVC)

Video Codecs, scalable content based streaming ap-

plications will become broadly adopted.

Research on scalable multimedia streaming has so

far gravitated on the use of MANE (Media Aware

Network Element) entities which fundamentally are

additional intermediate nodes used for manipulating

and customizing streaming sessions, in clear contrast

Urteaga I., Unanue I., Del Ser J., Sanchez P. and Rodriguez A. (2010).

ON THE DESIGN OF A SCALABLE MULTIMEDIA STREAMING SYSTEM BASED ON RECEIVER-DRIVEN FLOW AND CONGESTION AWARENESS.

In Proceedings of the International Conference on Signal Processing and Multimedia Applications, pages 39-45

DOI: 10.5220/0002987400390045

 SciTePress

to client-server based end-to-end services. This ap-

proach has been thoroughly analyzed and studied in

the literature. For instance, in (ASTRALS, 2010;

Schierl et al., 2007; Renzi et al., 2008) N RTP (Real

Time Protocol) sessions transmitted by the server are

fused by the MANE into a single RTP ﬂow for each

client according to network conditions. Still, other

contributions (Liebl et al., 2006; Tizon and Pesquet-

Popescu, 2008) propose to use MANEs to perform

an optimized packet scheduling and radio resource

sharing over the last wireless hop of a network by

mapping scalable content layer dependencies to ﬂow

priorities. Unfortunately, the aforementioned use of

MANEs presents several disadvantages: 1) the inser-

tion of an intermediate media-aware device into the

streaming scenario, and 2) the need for modifying

both RTP and RTCP (Real Time Control Protocol)

packets to adapt them to the customized content. De-

ploying MANEs into the streaming system requires to

know beforehand where the ﬁnal clients are located.

Since the success of streaming services is only

achievable if respecting the self-regulatory nature of

transmissions within the Internet, it is mandatory to

avoid either overloading or under-utilizing network

resources. This justiﬁes the need for providing con-

gestion control techniques. Several congestion con-

trol mechanisms have been presented for streaming

applications in the literature, e.g. see (Feamster et al.,

2001; Ma and Ooi, 2007; Mujica-V. et al., 2004;

Papadimitriou and Tsaoussidis, 2007) and references

therein. However, most of them are source-based

(i.e. the transmitting node is in charge for implement-

ing congestion-aware techniques), which requires ac-

tive probing, information piggybacking or acknowl-

edgement mechanisms. Several drawbacks can be in-

ferred from the application of source-based conges-

tion protocols to heterogeneous IP networks. On one

hand, demanding feedback from the client implies

an overhead in terms of both processing complexity

at the client and bandwidth over-utilization. On the

other hand, in networks where both wired and wire-

less technologies coexist links with highly asymmet-

ric characteristics are likely to appear. Therefore, tak-

ing the two-way path into account is not desirable

in heterogeneous networks, since asymmetric charac-

teristics of paths cannot be reliably estimated at the

server side. Thus, worst conditions prevail in two-

way path congestion control, as it is not possible to

distinguish whether the problem arises in the uplink

or the downlink. The heterogeneous nature of future

networks implies the need for new receiver-driven

congestion control mechanisms. We here propose to

discard considering two-way paths and, in-contrast, to

only account for the down-link state in our receiver-

driven network congestion.

This position paper outlines the key design prin-

ciples of a receiver-driven streaming system based on

scalable multimedia content. Both the management

of the multimedia content and the congestion and ﬂow

control logic are placed on the client, hence minimiz-

ing the computational complexity of the server. In

our approach, not only streaming standards are kept

unmodiﬁed, but we also proﬁt from the information

already embedded by such protocols. Furthermore,

each constituent component of our proposed architec-

ture is independent from each other. Another novel

contribution of our work hinges on the metrics uti-

lized for the congestion and ﬂow control mechanisms,

for which we introduce a novel LER (Loss Event

Rate) metric which is proven to offer enhanced stabil-

ity to bursty losses with respect to conventional packet

loss rate metrics.

The remainder of this manuscript is organized

as follows: ﬁrst Section 2 introduces the reader to

the fundamentals of scalable multimedia streaming

whereas Section 3 presents our novel receiver-driven

end-to-end streaming system proposal for distributing

scalable media content. Finally, concluding remarks

and future research lines are drawn in Section 4.

2 SCALABLE STREAMING

Due to the heterogeneity of the actual networks and

the proliferation of a wide range of ﬁnal devices, it

is essential to adapt the streaming content for each

speciﬁc context. Early approaches have been based

on storing a number of replicas of the same origi-

nal content or, alternately, on transcoding the original

content in a case-by-case basis. Recently, research

efforts have been conducted towards the generation

of inherently scalable multimedia content as a means

to provide different versions of the same multime-

dia content, without resorting to multiple successive

transcoding tasks. Consequently, processing redun-

dancy and storage occupancy of the encoded multi-

media content are minimized.

This growing interest in scalable codiﬁcation has

led to several research lines: the SVC (H.264/SVC,

2009) and MVC (H.264/MVC, 2009) extensions of

the so-called Advanced Video Codec (H.264/AVC).

The H.264/SVC standard attains high compression

rates while simultaneously combining three scalabil-

ity levels into a single encoded bitstream, namely spa-

tial (resolution), temporal (frame rate) and signal-to-

noise ratio (SNR) scalability (ﬁdelity). H.264/SVC

The authors recommend the reader to skip Section 2 if

familiar with the concepts tackled therein.

SIGMAP 2010 - International Conference on Signal Processing and Multimedia Applications

encodes a given video content to a layered structure

consisting of a base layer (comprising the lower lev-

els of each of the mentioned scalabilities) and a num-

ber of enhancement layers. The goal of the enhance-

ment layers is to progressively reﬁne (in terms of each

aforementioned scalabilities) the base layer so as to

obtain better end-user Quality of Experience (QoE)

degrees. The H.264/MVC video codec (currently un-

der development) is also based on a layered bitstream

structure. However, it additionally provides multiple

views of the same scene which allows rendering 3D

perspectives. Thanks to the scalability of this codec,

it is possible to choose a speciﬁc viewpoint of a scene,

while keeping high encoding efﬁciency through inter-

view predictions.

Additionally, the Internet Streaming Media Al-

liance (ISMA, 2010) promotes the use of certain stan-

dard protocols for streaming applications: Real Time

Streaming Protocol (RTSP), Session Description Pro-

tocol (SDP), Real Time Protocol (RTP) and Real Time

Control Protocol (RTCP). RTSP is responsible for es-

tablishing and controlling the streaming sessions in

real time. SDP describes the whole streaming ses-

sion, as it characterizes the content and the stream-

ing session itself. RTP delivers the multimedia con-

tent to destination in combination with RTCP, which

communicates statistics and control information for

each RTP session. Finally, it is important to recall

that standardization organisms periodically evaluate

and update the set of recommended standards to meet

the requirements and constraints of newly designed

multimedia codecs.

3 SYSTEM DESIGN

A block diagram of the proposed end-to-end stream-

ing system design is depicted in Figure 1. As shown,

the ﬁrst processing steps of the streaming system con-

sist basically of multimedia content encoding and

encapsulation. Without loss of generality we have

hereby adopted the H.264/SVC codec for ensuring

scalability at the encoding process. The only require-

ment imposed by our design is that real-time layer

switching must be supported during the decoding pro-

cess. As for the encapsulation, the MP4 ﬁle for-

mat has been selected due to 1) the H.264/SVC spe-

ciﬁc extension (AVC, 2008) included in such stan-

dard, and 2) the so-called hint tracks. Hint tracks en-

able a media-unaware streaming server by indicating

how to perform the streaming disregarding the con-

tent itself, thus alleviating the server from the compu-

tational burden derived from analyzing the streaming

peculiarities of each speciﬁc content.

Our system performs a multilayered streaming

where the M layers conforming the scalable content

are mapped to N RTP sessions {RT P

}

N−1

i=0

. Observe

that even if the end-user appreciates a single multi-

media stream at reproduction, the content is received

in n ≤ N parallel RTP sessions, where n denotes the

actual number of demanded RT P

by a given client.

As deﬁned in (RTP, 2010), the mapping between SVC

layers and RT P

can be done by following distinct cri-

teria. In our system the mapping rule is provided to

the server through hint tracks in the encapsulation.

Figure 1: Block diagram of the proposed end-to-end scal-

able content streaming system.

Thanks to multilayered streaming, the character-

istics of received multimedia content are dependent

on the actual number of transmitted RT P

and there-

fore, diverse end-user requirements can be easily met.

Our system design allows each client to select the

subset of RT P

that better fulﬁlls its needs, as de-

scribed in Subsection 3.1. Furthermore, the client

performs a receiver-driven congestion and ﬂow con-

trol by adapting n, based on both network and client

conditions (Subsection 3.2). The justiﬁcation for

this receiver-driven approach is to avoid complex and

highly-loaded servers by balancing the computational

load between clients. Besides, piggybacking other-

wise necessary client information and network state

parameters to the server is circumvented. Finally,

it should be clear that sharp and frequent transitions

among video layers are extremely displeasing for the

QoE. In such situation a smoother video of reduced

bit rate is then preferred rather than an inconsistent

and jerky high quality video. In Subsection 3.2.2 we

outline several criteria to achieve smooth multimedia

reproduction aimed at maintaining a satisfactory QoE.

3.1 Content Streaming Procedure

In our proposal the client is the unique responsible

for (throughout the whole streaming session) dynam-

ically controlling n, i.e. the number of scalable multi-

media layers to be received. We remark that our sys-

tem follows IETF’s speciﬁcations concerning scalable

content over streaming protocols (RTP, 2010). The

presented streaming process begins with the client de-

manding information to the server about some spe-

ciﬁc multimedia content by sending a RTSP DE-

SCRIBE request. The server responds to the client

ON THE DESIGN OF A SCALABLE MULTIMEDIA STREAMING SYSTEM BASED ON RECEIVER-DRIVEN

FLOW AND CONGESTION AWARENESS

with the SDP description of the required content over

RTSP. This SDP description contains all the infor-

mation regarding that particular multimedia stream-

ing session: number and characteristics of each RT P

that conform the streaming session, dependencies be-

tween different RT P

, and so on. At this point

the client is capable of selecting a subset of RT P

depending on its processing and memory capabili-

ties. Once this is set, the client triggers the stream-

ing process by sending the RTSP SETUP and RTSP

PLAY commands to the server. Throughout the whole

streaming session, the client is able to cancel or de-

mand (using RTSP commands) each RT P

described

in the SDP, as long as dependencies among scalable

content layers are met. Once the content is received in

the client, the RTP packets corresponding to different

RT P

are merged and ordered in a single bitstream,

which is next depacketized and decoded. Finally, the

multimedia content is displayed.

The determination of the optimal number of scal-

able layers and their mapping to RTP

sessions is both

application and content dependent, however our sys-

tem is generically designed (independent from spe-

ciﬁc mappings). Hence, we assume that RT P

ses-

sions (and, consequently, scalable layers) are cor-

rectly ordered beforehand in the encapsulation pro-

cess, so the user simply needs to comply with the in-

formation provided by the SDP. In order to maximize

the QoE, we propose a soft and stable layer switching

mechanism further detailed in Subsection 3.2.2.

3.2 Flow and Congestion Control

In IP networks, several trafﬁc types and ﬂows com-

pete for the available scarce resources, which re-

quires avoiding either trafﬁc overload or the under-

utilization of the network resources. Congestion can

be induced by both attempting to oversubscribe the

processing capabilities of intermediate nodes or by

over-demanding network link capabilities. This ra-

tionale, along with the diverse memory and process-

ing characteristics presented by end clients, motivates

the need for appropriate congestion and ﬂow control

mechanisms in streaming systems. However, multi-

layered streaming imposes several considerations to

be taken into account. First, both congestion and ﬂow

control must be based not only on a single ﬂow, but on

several parallel RT P

. Second, each RT P

has a ﬁxed

transmission bit rate enforced by the scalable content

requirements. Thus, RT P

by themselves cannot ex-

pand nor reduce their bandwidth usage.

Consequently, congestion and ﬂow control for

multilayered streaming can only be accomplished

based on discrete bitrate intervals. This certainly

poses several design challenges gravitating on the

tradeoff between reactivity to network and client dy-

namic characteristics (which justiﬁes relatively short

control periods) and the QoE degradation due to layer

switching. We intend to balance this tradeoff by ben-

eﬁting from the speciﬁc features of scalable content

streaming, which gives rise to a novel receiver-driven

congestion and ﬂow control mechanism.

3.2.1 Metrics

The proposed metrics are restricted to the information

available at the client side. Therefore, procedures

such as message piggybacking or probing (e.g. for

bandwidth estimation) are discarded. Network state is

inferred by extracting information from the received

RT P

packets, while reception buffers are monitored

for estimating the client state. The following metrics

will be sampled and computed for each received

RT P

(i ∈ {0,..., n − 1}) every T

seconds which, at

the early stage of this research, is believed to be a

multiple of the GOP (Group of Pictures) size:

A) Buffer state, B

: it quantiﬁes the load at the re-

ceiver for each received RTP

. Let b

∈ [0, 1] denote

the buffer state of session RT P

at time t. At this

early stage of our research we deﬁne B

= Γ(b

,∆b

) ∈

[−1,1], where ∆b

= b

− b

t−T

. Γ(·) is a monotoni-

cally increasing function with its two parameters. No-

tice that this generic deﬁnition of B

not only accounts

for the current state of the buffer, but also accommo-

dates sharp changes on it.

B) Interarrival jitter, J

: the interarrival jitter is de-

ﬁned as the mean deviation of the difference (D) in

packet spacing at the receiver compared to the sender

for a pair of packets. In our case, computation is done

based solely on the timestamp values of received RTP

packets. Let S

and R

denote the timestamps for the

p-th RTP packet at transmission and reception, re-

spectively. The packet spacing difference at session

RT P

for packets p and q will be given by

(p,q)

= (R

-R

)-(S

-S

) = (R

-S

)-(R

-S

). (1)

The interarrival jitter for the received packet p within

session RT P

, denoted as j

, is given by

= j

p−1

+ (|D

p−1,p

| − j

p−1

)/16, (2)

and the continuous interarrival jitter at time t for

RT P

, denoted as j

, will be set equal to the interar-

rival jitter j

of the last received packet for each RT P

It should be noted that j

is continuously updated

upon reception of each RTP packet. Then, every T

seconds the overall Interarrival jitter J

is computed

as J

= Ψ( j

,∆ j

) ∈ [−1,1], where ∆ j

= j

− j

t−T

SIGMAP 2010 - International Conference on Signal Processing and Multimedia Applications

and j

∈ [0, j

MAX

], with j

MAX

denoting the maximum

permissible delay for packet decoding. Similar to

Γ(·), Ψ(·) is a monotonically increasing function with

its two parameters.

C) Loss Event Rate, LER

: it deﬁnes the rate at which

packet loss events occur. Although the fraction be-

tween sent and received packets is typically used as

a congestion indicator, it does present several draw-

backs. When bursty losses occur, the value of such

fraction metric decreases sharply from which serious

network congestion is deduced. However, successive

losses do not necessarily involve severe congestion,

specially in wireless communications subject to inter-

ference and collisions. To overcome this issue, the

novel packet Loss Event Rate LER

metric proposed

here comprises both isolated and bursty losses within

a predetermined evaluation interval T

eval

. In other

words, LER

is the frequency of packet loss events

(either single or multiple) during a T

eval

period for

each RT P

measured at time t = kT

eval

To compute this metric, every T

eval

seconds the

client detects any packet loss based exclusively on

checking the sequence number information provided

by incoming RTP packets. Two variables are pro-

gressively updated after the loss detection process:

last

and I

new

. The ﬁrst refers to the index of the

last evaluation period with either bursty or isolated

packet losses, whereas the second is updated to the

current evaluating interval index if any packet loss is

detected. Based on these two variables, the instan-

taneous loss event rate ILER

for the k-th evaluation

interval is computed as

ILER

(

0 if no packet loss detected,

new

-I

last

otherwise,

(3)

from which a global weighted Loss Event Rate LER

is recursively computed every T

eval

seconds as

LER

= δ · ILER

bt/T

eval

+ (1 − δ) · LER

t−T

eval

. (4)

In the above deﬁnition, δ ∈ (0,1] is an arbitrary

parameter that trades exhaustive traceability of the

packet losses (δ = 1) for the smooth estimation of

the packet loss trend (δ → 0). It is also assumed that

LER

∈ [0,1]: if no losses occur, LER

= 0 and, oth-

erwise, if every T

eval

any packet is lost, LER

= 1.

3.2.2 Decision Criteria

The congestion and ﬂow control mechanism builds

upon the above deﬁned B

(ﬂow), J

and LER

(con-

gestion) metrics. In fact, J

is a signiﬁcant indicator

for initial network congestion. When the network is

unable to correctly process trafﬁc data, the packet de-

lay increases even in absence of packet losses. When

network congestion increases further, packet losses

occur as the LER

metric would reﬂect.

As multilayered streaming is considered in our

system design, the whole set of {RT P

}

n−1

i=0

must be

considered at the receiver. However, note that each

session does not have the same relevance due to the

dependencies between scalable content layers (e.g. as

RT P

contains the base layer, such session must be

given full processing priority). Thereby, every t = kT

seconds a set of accumulated metrics (B

,LER

)

for the n RT P

sessions is obtained by applying dif-

ferent weights α

, namely

n−1

∑

i=0

· B

(Accumulated Buffer State), (5)

n−1

∑

i=0

· J

(Accumulated Jitter), (6)

LER

n−1

∑

i=0

· LER

(Accumulated LER). (7)

It should be clear that since session RT P

contains the

base layer, max{α

}

n−1

i=0

= α

. Also observe that the

values of the weights for the three metrics within a

given session index are set equal. Nevertheless, bal-

ancing the importance between B

, J

and LER

accomplished by utilizing different coefﬁcients in the

metric fusion stage, which merges the above accumu-

lated metrics into an overall ﬂow-congestion indicator

=Ω(

,LER

)=γ

·B

+γ

·J

+γ

·LER

, (8)

where it should be remarked that Ω(·) can be set to

any other (not necessarily linear) combination of the

accumulated metrics. Finally, a decision rule is taken

every T

based on ζ

. The decision logic determines

whether a new RT P

can be demanded from the server

(i.e. n is increased to n +1) without degrading the per-

formance of both client and network, or if it is instead

mandatory to reduce the number of sessions received

(n = n − 1), i.e.

n =



n + 1 if ζ

< ζ

(n),

n − 1 if ζ

> ζ

(n).

(9)

Note that decision limits ζ

(n) (add) and ζ

(n) (re-

move) are not static values but depend on the num-

ber of RT P

sessions received by the client(n). By

following this approach inter-session fairness is guar-

anteed, since we facilitate the demand of new RT P

for low-quality streams, while restraining high quality

streams from demanding more RT P

sessions. There-

fore, ζ

(n) is a monotonically increasing function

ON THE DESIGN OF A SCALABLE MULTIMEDIA STREAMING SYSTEM BASED ON RECEIVER-DRIVEN

FLOW AND CONGESTION AWARENESS

with n, bounded in the range [0 + ε

,1 − ε

], while

(n) is a monotonically decreasing function with n

with support [0 + ε

,1 − ε

], where all ε’s are design

parameters. Unfortunately, our decision logic still

poses the hazard of entering an unstable state when

iterating between adjacent RT P

. Since frequent layer

switching degrades the QoE at content reproduction,

we propose a safeguard mechanism: only if the sta-

bility of the system (based on proposed both network

and client metrics) is guaranteed during a predeter-

mined interval T

, the scalable layers contained in the

newly received RT P

are served to the decoder and ﬁ-

nally, delivered to the end-user.

The deﬁnition of the Ω(·), Ψ(·) and Γ(·) func-

tions, as well as the obtention of optimum values for

the decision limits ζ

(n),ζ

(n) and the intervals T

and T

are not straightforward. In order to perform a

satisfactory receiver-driven ﬂow and congestion con-

trol, the following guidelines should be met:

• The receiver-driven control procedure should be

responsive to sudden changes on any of the above

metrics, and allow RT P

session dropping as the

value of any of such metrics becomes critical, i.e.

→ 1, J

→ 1 or LER

→ 1.

• The decision rule must be specially sensitive to

the buffer state, as it dominates client’s perfor-

mance even in absence of network congestion.

• Iterating between adjacent RT P

should be cir-

cumvented to avoid continuous layer switching

which in turn degrades QoE.

• Inter-session fairness should be achieved. It is

preferable to have equal-quality multimedia ﬂows

than streaming sessions with strongly asymmetric

quality levels.

4 FUTURE RESEARCH

In this paper we have presented a novel end-to-end

scalable content based streaming system aimed at

maximizing the end-user’s QoE. Our system prof-

its from the virtues of the scalable content to per-

form a multilayered streaming, where each client is

able to retrieve a personalized content. Being scal-

able encoding the only limitation imposed to our sys-

tem, we determine to keep intact the involved stream-

ing standards and maximize system component inde-

pendence. Furthermore, due to our receiver-driven

congestion and ﬂow control algorithm, the streaming

session is adapted to both dynamic changes in net-

work’s state and to client’s limitations. Our presented

control metrics (Buffer State, Interarrival Jitter and

Loss Event Rate) are restricted to information already

available in streaming clients. Besides, recall that the

Loss Event Rate has been speciﬁcally designed to im-

prove congestion control performance over heteroge-

neous networks.

Further investigation will be conducted towards

the deﬁnition of the weights α

, the Ω(·), Ψ(·) and

Γ(·) functions, and the decision limits ζ

(n) and

(n). To this end, a threefold criteria will be adopted:

1) to be responsive to both sudden changes and crit-

ical values of network’s and client’s state metrics; 2)

to emphasize on client’s buffer state during the con-

trol procedure; and 3) to ensure inter-session fairness

among the streaming clients.

REFERENCES

ASTRALS (2010). ASTRALS Project (FP6-IST-028097):

http://www.ist-astrals.org/.

AVC (2008). File format support for Scalable Video

Coding, ”Information technology - Coding of audio-

visual objetcs - Part 15: Advanced Video Cod-

ing (AVC) ﬁle format AMENDMENT 2”, ISO/IEC

14496-15:2004/AMD2:2008(E).

Feamster, N., Bansal, D., and Balakrishnan, H. (2001). ”On

the Interactions Between Layered Quality Adaptation

and Congestion Control for Streaming Video”. In 11th

International Packet Video Workshop.

H.264/MVC (2009). H.264/MVC, ”Ammendment H of In-

formation technology - Coding of audio-visual objects

- Part 10: Advanced video coding”, ISO/IEC 14496-

10:2009(E).

H.264/SVC (2009). H.264/SVC, ”Ammendment G of In-

formation technology - Coding of audio-visual objects

- Part 10: Advanced video coding”, ISO/IEC 14496-

10:2009(E).

ISMA (2010). Internet Streaming Media Alliance:

http://www.isma.tv/.

Liebl, G., Schierl, T., Wiegand, T., and Stockhammer, T.

(2006). ”Advanced Wireless Multiuser Video Stream-

ing using the Scalable Video Coding Extensions of

H.264/MPEG4-AVC”. In ICME, pages 625–628.

IEEE.

Ma, L. and Ooi, W. T. (2007). ”Congestion Control in Dis-

tributed Media Streaming”. In 26th Annual IEEE Con-

ference on Computer Communications (INFOCOM

2007).

Mujica-V., V. E., Sisalem, D., Popescu-Zeletin, R., and

Wolisz, A. (2004). ”TCP-Friendly Congestion Con-

trol over Wireless Networks”. In European Wireless

2004.

Papadimitriou, P. and Tsaoussidis, V. (2007). ”SSVP:

A Congestion Control Scheme for Real-time Video

Streaming”. Computer Networks, 51(15):4377–4395.

Renzi, D., Amon, P., and Battista, S. (2008). ”Video Con-

tent Adaptation Based on SVC and Associated RTP

SIGMAP 2010 - International Conference on Signal Processing and Multimedia Applications

Packet Loss Detection and Signaling”. In Ninth Inter-

national Workshop on Image Analysis for Multimedia

Interactive Services, pages 97–100.

RTP (2010). RTP Payload Format for SVC

Video (”draft-ietf-avt-rtp-svc-20.txt”)

http://tools.ietf.org/html/draft-ietf-avt-rtp-svc-20.

Schierl, T., Hellge, C., Mirta, S., Gr

uneberg, K., and Wie-

gand, T. (2007). ”Using H.264/AVC-based Scalable

Video Coding (SVC) for Real Time Streaming in

Wireless IP Networks”. In IEEE International Sym-

posium on Circuits and Systems (ISCAS).

Tizon, N. and Pesquet-Popescu, B. (2008). ”Scalable and

Media Aware Adaptive Video Streaming over Wire-

less Networks”. EURASIP Journal on Advances in

Signal Processing, 2008(168).

ON THE DESIGN OF A SCALABLE MULTIMEDIA STREAMING SYSTEM BASED ON RECEIVER-DRIVEN

FLOW AND CONGESTION AWARENESS