World Academy of Science, Engineering and Technology 46 2008
Performance Evaluation of Issues Related to
Video over Broadband Network
A. S. Gundale, and A. R. Yardi
was designed to dynamically adapt to properties of the
internetwork and to be robust during various network
parameter failures. TCP was formally defined in RFC 793.
As time went on, various errors and inconsistencies were
detected, and the requirements were changed in some areas.
These clarifications and some bug fixes are detailed in RFC
1122. Extensions are given in RFC 1323. Each machine
supporting TCP has a TCP transport entity, either a library
procedure, a user process, or part of the kernel. In all cases,
it manages TCP streams and interfaces to the IP layer.
A TCP entity accepts user data streams from local
processes breaks them up into segments not exceeding 64
KB and sends each piece as a separate IP datagram. When
datagrams containing TCP data arrive at a machine, they are
given to the TCP entity, which reconstructs the original byte
streams. These segments supplied to IP layer converts them
in to IP datagram. The IP layer gives no guarantee that
datagrams will be delivered properly, so it is up to TCP to
time out and retransmit them if datagram is corrupted or not
received. Datagrams that do arrive may well do so in the
wrong order; it is also up to TCP to reassemble them into
messages in the proper sequence. In short, TCP must furnish
the reliability that most users want and that IP does not
provide. The TCP/IP networking protocol (Internet and
World Wide Web rely on) shows poor performance over
satellite network. The TCP/IP reference implementation’s
4k buffer size limits channel capacity and the data
throughput to only 64kbps. The maximum buffer of 64k
limits maximum throughput to 1Gbps. Therefore the GEO
‘K’ band satellite services are unsuitable for high bandwidth
application as the increased GEO connection decrease the
available bandwidth. With a limited buffer size, a longer
end-to-end delay decreases the space available to hold
copies of unacknowledged data for retransmissions. This
limits the throughput on a loss-less TCP connection.
The TCP is accompanied by congestion control
algorithm, which is forced to react during loss of packets
even though the packet losses are not due to congested
network. When the loss is indicated by a timeout, TCP
enters into a slow-start phase and throughput reduces
significantly. Duplicate acknowledgements forces system to
go into fast recovery mode. This results in reduction of the
congestion window size. On the other hand, whenever a
packet loss is due to a transmission error, any contraction of
the window is inappropriate. The outcome is decreased
throughput, poor utilization of the network resources.
Therefore, modifications of TCP to mitigate the conditions
imposed by wireless and mobile networks are needed.
Obviously, there is no a single solution for making TCP to
Abstract—Multimedia
applications
such
as
video
conferencing, video on demand, teleteaching etc are the major
users of broadband networks. At the heart of this revolution is the
digital compression of audio and video signals. Network
supporting multimedia applications require certain performance
guarantees from networks. Moreover when these applications are
transmitted over satellite network, issues like limited bandwidth
and throughput restrict them. Efficient allocation techniques are
needed to provide guaranteed transfer. Various methods such as
TCP and IP are available to transport multimedia applications. The
TCP shows degradation in its performance in typical satellite
environment due to slow-start, window size, and acknowledgment
policy. Also there were attempts to deliver applications over IP in
satellite, but leads to wastage of expensive network capacity. The
Asynchronous Transfer Mode network is designed to support both
real and non-real time broadband applications making it very
suitable for transporting multimedia. This paper discusses issues
related resource management over satellite networks.
Keywords—ATM, cwnd, Mpeg, QoS, VoD.
M
I. INTRODUCTION
OST of the networks using TCP/IP are developed
for data transportation. TCP was designed to run
over any short distance packet switching wired networks
where errors due to the transmission medium. TCP adopts
additional control mechanisms like flow control, congestion
management and reliability to provide efficient network
utilization and a fair share of network resources. TCP in
wireless and mobile networks suffers significant
performance degradation due to the misinterpretation of
random losses that come from high bit-error rate, limited &
variable bandwidth and frequent disconnections. Generally
a network experiences two kinds of traffics, one is elastic –
those adjust themselves in accordance with network
parameter changes e. g. FTP & HTTP and another is
inelastic- which do not follow changes in network
parameters like delay or throughput. This applies mostly to
real-time traffic, which imposes tight restrictions on QoS
parameters such as throughput, delay, jitter and packet loss.
Attempts are also made to deliver IP over satellite network,
but the satellite technologies have focused on connectionoriented transmission protocols rather than IP.
II. TRANSMISSION CONTROL PROTOCOL (TCP)
TCP was specifically designed to provide a reliable endto end byte stream over an unreliable internetwork. TCP
A. S. Gundale is withWalchand Institute of Technology, Solapur, India.
A. R. Yardi is with Walchand College of Engineering, Sangli, India.
122
World Academy of Science, Engineering and Technology 46 2008
operate flawlessly in heterogeneous networks.
There are various mechanisms are available to control the
congestion. Some of the popular methods are TCP Reno,
TCP FACK, TCP Westwood and TCP VEGAS etc. The
Forward Acknowledgment (FACK) algorithm improves
on the recovery of the lost packets. It uses SACK (Selective
ACK) options in order to gather additional information
about the state of congestion and provides more accurate
control to the outstanding data in the network. FACK
decouples the congestion control algorithms from the data
recovery algorithms. Namely, new data can be sent during
network recovery of the lost data in order to sustain the TCP
self-clocking feature when there is no further data to
retransmit.
FACK Algorithm1. For more accurate calculation of outstanding packets
2. awnd = snd.nxt – snd.fack
with the assumption of no out-of-order delivery and
no retransmission
3. To include retransmission,
awnd = snd.nxt – snd.fack + retran_data
4. While (awnd < cwnd) sendsomething ();
One of the strong features of ATM is its quality of service
(QoS) guarantees like delay, cell loss and bandwidth
guarantees. Real-time voice and video (and applications
such as VoD) require stringent delay and bandwidth
guarantees for acceptable picture quality, which makes
ATM very suitable for carrying multimedia traffic.
Each application expects the network to provide a desired
quality of service (QoS). QoS measurements include bounds
on the cell loss probability, cell delay, etc. The service
provider is interested in providing the desired QoS
efficiently. For Conventional approaches of resource
allocation rely on predetermined traffic characteristics. The
amount of resources required to provide the QoS is
calculated using these values. These techniques experience
the following fundamental problems. First, the source
characteristics may not be known ahead of time. In the case
of live or interactive video, the user must guess at these
characteristics. Second, parameters may not adequately
characterize the source. It has been shown for MPEGcompressed video that long-range dependencies occur,
which implies that standard statistical models are probably
inadequate. Third, the number of parameters required
should be kept small, so to reduce the complexity of the
allocation method.
IV. PERFORMANCE EVALUATION OF MPEG OVER ATM
Network considered for simulation is as shown in Fig. 2.
The model composed of an ATM switch with a finite
capacity buffer, bottleneck link (L) and three VBR MPEG
connections .The distance between the sources and the
switch (L) is adjustable). The Round Trip Time (RTT)
between the sources and destinations is also variable.
Fig. 1 Performance of FACK over satellite network
Source 1
The above simulation shows congestion control during
FTP over GEO satellite network with BER 5%. The
performance shows frequent degradation in throughput, not
necessarily because of congested link but may be due to
frequent disconnections. In this control mechanism more
time is wasted in reestablishing the connection and
regaining the maximum throughput conditions.
2
CBR
VBR
ABR
2
Sink 3
Each Mpeg source is considered as an independent
process, which wakes up periodically to send a frame. Each
cell corresponding to a frame is send to buffer separately
and all cells belonging to a frame has to be sent to the buffer
before the next frame from source is generated. As this
process is periodic, this period is divided into equal time
slots depending on frame size. After all cells received in
buffer, it is sent them immediately on the channel if it is idle
and link is set to busy. If link is busy then cell is put into the
queue. Whereas if link is busy & buffer is full, then
incoming cell is discarded and related statistics is updated.
Buffer size decides maximum number of cells to be queued.
Additional request is rejected and corresponding cell is
dropped.
1. Analysis of Effect of finite buffer size: The queuing
behavior of network buffer is analysed. As many Mpeg
streams share the buffer, the cell loss due to limited buffer
size is important consideration. The simulation is carried for
number of Mpeg sources on the input side at different
network utilization conditions.
TABLE I
CHARACTERISTICS OF ATM SERVICES
QoS
Guarantee
Yes
Yes
Acceptable
BTE
Fig. 2 Simulation model
ATM is a totally different transfer technology from the
traditional Synchronous Transfer Mode (STM) technology.
ATM is based on statistical multiplexing which allows
transportation of real-time and non-real time services in the
same network. ATM Forum has defined and standardized
four types of services: Constant Bit Rate (CBR), Variable
Bit Rate (VBR) including real time VBR and non-real time
VBR, Available Bit Rate (ABR) and Unspecified Bit Rate
(UBR). Their specifications and characteristics are shown in
Table I.
BW
guarantee
Yes
Yes
Limited
Buffer
Source-3
III. ASYNCHRONOUS TRANSFER MODE
Service
Sink 1
BW
utilization
Very low
Low
Very high
123
World Academy of Science, Engineering and Technology 46 2008
It is observed that the more buffer size reduces the cell
loss whereas the more link utilization requirements results
in increased cell loss. Mpeg trace and simulated trace was
considered during simulation. Fig. 3 shows cell loss for six
multiplexed Mpeg sources at different network efficiency
conditions. The error rate was considered in the range of 101
to 10-3.
The average time is function of queue length- as queue
size is increased cell has to wait more. The average waiting
time for a cell in the queue is 50 mS, for two Mpeg
applications on the input and frame is generated periodically
at the rate one fame per 1/30 sec This waiting time is very
important as it contribute in QoS.
4. Average buffer requirement in erroneous conditions:
The error on the channel is responsible for cell loss.
Excessive cell loss during transmission couldn’t meet the
QoS requirements. It is observed that when cell loss rate is
10-4, the link utilization is less than 80% in two Mpeg
applications. It is also found that average buffer requirement
decreases as more applications are attached on the link. Fig.
6 shows average buffer requirement for six and ten Mpeg
input streams. It is experienced that 10 Mpeg applications
require less than 6 applications.
Fig. 3 Effect of buffer size on cell loss
2. Average Buffer behavior analyses: As cell is queued,
which are scheduled to be transferred, it has to wait before
is actually coupled to the link. The queue is managed in real
time i.e. when one cell is scheduled to transmit; the queue
size is increased by one, whereas after sending one cell on
the link the queue size is decremented by one. It necessary
to analyse the queue in case of Mpeg as it is bursty in
nature. This queue requirement is taken as average buffer
requirement. This is also analysed at various network
utilization requirements.
Fig. 6 Buffer requirement for different link utilization
Fig: 7 Cell loss against average buffer for different ink efficiency
requirements. Cell loss rate 10-4
Fig. 4 Average queue length versus cell loss for six multiplexed
sources
5. Effect of modeling parameters on cell loss: The Mpeg
consist of I, B and P frames. These frames are characterized
by various modeling parameters namely μI, σI, μB, σB, μP, σP.
The value of μI dose increase mean and variance of I frame
and leads to greater cell loss in case of bigger I frame. μI
doesn’t affect values of μB & μP.
3. Waiting time for a cell: Average waiting time depicts
the cell need to wait before it is transmitted. Careful
selection in necessary as for most of time it remains empty.
In case of Mpeg queue is holding some cells during big Iframe transmission. The increased link utilization also
demands more average queue.
Fig. 5 Average waiting time in the queue
Fig. 8 Effect of μI on cell loss
124
World Academy of Science, Engineering and Technology 46 2008
The parameter σI also shows approximately same
variation. In case of B frame μB & σB are interrelated. As μB
increases it forces to increase size of B frame in number of
cells. The increase in μB shows reduction in cell loss
percentage in case of both B and I frames. The parameter μP
is optimized, upto the optimized value the cell loss is within
the specified limits, but if value goes above predetermined
value the cell loss increases sharply. Hence with the
observations the cell loss depends on frame size and number
of frames. The autocorrelation factor a1 and a2 doesn’t
create much change in cell loss.
7. Forwarding ATM on broadband satellite network: For
long distance communication the use of satellite is
necessary. Hence ATM transmission on long BDP
(Bandwidth Delay Product) network is key. The satellite
network is using broadband terminal equipment along with
all the parameter with the local applications, as discussed
earlier. BTE also manages its own queue and its selection is
also based on required quality of services.
The values for buffer and queue are exerting their effects
on throughput on the channel. Up to certain value of queue
size the performance is optimized and thereafter
performance degrades.
Fig. 9 Effect of σI cell loss
6. Analysis of bandwidth utilization: The link efficiency
greatly depends on how the input application streams are
attached to the network. In previous simulations the input
streams were randomly forwarded. It is been very
encouraging to find that if input applications are attached
with some logical arrangement rather than randomly
connected, increase in link utilization is observed.
Considering simple superposed arrangement as given
belowu* = (0, 1, 2,…., L) then (0, 1, 2,…., L-1) then….. (0, 1, 2,
…, N-wL-1)
where w is integer part of N/L. The number u* indicates
application currently is in progress. The cell loss behavior is
analysed and it is found that the cell dropped are reduced.
Fig. 12 Retransmission of ATM cells for different buffer sizes
Fig. 13 Effect of BTE buffer size on throughput
V. CONCLUSION
1. Appropriate buffer size is essential to handle applications
on the link. As number MPEG sources are increased the
need of buffer gets minimized.
2. Buffer size depends on utilization expected and
permissible cell loss (QoS)
3. As sources get increased the average waiting time also
gets increased.
4. If sources are handled in specific multiplexed manner the
buffer requirement further gets reduced. Hence it is clear
that random handling of source is not beneficial.
5. As MPEG is bursty, an appropriate bandwidth must be
kept in reserve to serve extra need at times.
6. ATM on Satellite broadband network needs careful
design to support appropriate quality of service.
Fig. 10 Cell loss against buffer size in superposed scheme
REFERENCES
[1]
[2]
Fig: 11 Cell loss against buffer size in random distribution scheme
125
M.F. Alam, M. Atiquzzaman, M.A. Karim, Efficient MPEG video
traffic shaping for the next generation Internet, IEEE GLOBECOM’
99, Rio de Janeiro, Brazil December (1999).
M.F. Alam, M. Atiquzzaman, M.A. Karim, Effects of source traffic
sha ping on MPEG video transmission over next generation IP
network, Eighth IC3N, Boston, USA October (1999) 514–519.
World Academy of Science, Engineering and Technology 46 2008
[3]
[4]
[5]
[6]
[7]
[8]
ATM Forum Technical Committee, Video on demand specification
1.0, Technical Report, ATM Forum, December 1995.
D. Deloddere, W. Verbiest, H. Verhille, Interactive video on demand,
IEEE Communication Magazine 32 (5) (1994) 82–88.
Y.-H. Chang, D. Coggins, D. Pitt, An open-system approach to video
on demand, IEEE Communication Magazine 32 (5) (1994) 68–80.
R.J. Jones, Baseband and passband transport system for interactive
video on demand, IEEE Communication Magazine 32 (5) (1994) 90–
101.
S. Lederman, Video on demand: a traffic model and GOS technique,
GLOBE-COM’86 (1986) 676–683.
W.D. Sincoskie, Video on demand: is it feasible? GLOBECOM’90,
San Diego, CA (1990) 201–205.
126