CESNET Te

hni al Report 6/2002

In uen e of Network QoS Chara teristi s on

MPEG Video Transmission
Sven Ubik, ubik esnet. z
Josef Vojte h, xvojte jfel. vut. z
June 21, 2002

1 Experiment obje tive

One of the prospe tive appli ations expe ted to be used frequently in National Resear h and
Edu ational Networks (NRENs) is high-quality video transmission. Appli ations of this type
use mostly MPEG en oding for video and audio data (MPEG1, MPEG2 or MPEG4). Required
bandwidth, ranging approximately from 2 Mb/s to 10 Mb/s, is relatively small when ompared
to the ba kbone apa ity of urrent NRENs. However, these video transmissions are supposed
to be available also in points farther away from the ba kbone. A typi al example is broad asting
a onferen e session from a le ture hall onne ted using a wireless link with the apa ity in
the order of 10 Mb/s to 100 Mb/s having less than optimal network QoS hara teristi s. The
obje tive of this experiment was to verify in uen e of primary QoS hara teristi s (loss rate,
delay and jitter) on the quality of video and audio signal transmitted using MPEG en oding.

2 Test onguration
The test onguration is shown in Fig. 1. The sending PC was equipped with Optibase MPEG
MovieMaker 200 en oding ard and the re eiving PC was equipped with Optibase Videoplex
Xpress de oding ard. Both ards support MPEG1 and MPEG2 en odings in SIF, QSIF, FullD1 and Half-D1 formats. The Moviemaker 200 ard an transmit MPEG1 data en oded online
100BASE-TX

100BASE-TX
Optibase MPEG
MovieMaker 200

Linux + NIST Net

Figure 1: Test onguration

1

Optibase Videoplex Xpress

from the S-video input or it an read MPEG1 or MPEG2 data stored in a le. We used this
parti ular hardware be ause it was just what we managed to get for testing purposes.
The sending and the re eiving omputer ommuni ated over a router based on a Linux
ma hine running NIST Net [1 emulation pa kage. NIST Net allows to ongure network
QoS hara teristi s experien ed by tra passing the router. These hara teristi s in lude
adjustable throughput, loss rate, dupli ations and delay (in luding standard deviation and
linear orrelation). The user an supply his own distribution fun tion for delay emulation, if
required.

3 Using NIST Net emulation pa kage

Using the NIST Net emulation pa kage involves three steps: loading the nistnet kernel module,
turning the NIST Net on and onguring the QoS hara teristi to be emulated.
To load the nistnet kernel module, run the Load.Nistnet s ript. In most ases you an
also just type insmod nistnet.
To turn NIST Net on, use nistnet -u (up) ommand.
To ongure the QoS hara teristi s to be emulated, use nistnet ommand with -a (add)
option, spe ifying sour e and destination IP addresses of the ow (optionally in luding proto ol
and port numbers) and required QoS hara teristi s. For example:
nistnet -a 195.113.147.1 195.113.147.2 --drop 0.1 --delay 100/10 --dup 0.1

This ommand will insert pa ket loss rate of 0.1%, mean delay of 100 ms with standard
deviation of 10 ms and dupli ations of 0.1% into the ow of pa kets sent from IP address
195.113.147.1 to IP address 195.113.147.2.

4 Observations
We found the following observations to apply almost equally to dierent stream types (MPEG1
SIF, MPEG1 QSIF and MPEG2 Half D1) and rates ranging from 4 Mb/s to 10 Mb/s.
As expe ted, pa ket loss rate was a riti al parameter. MPEG without FEC proved to be
ompletely intolerable to pa ket losses as indi ated in Table 1. With 10 Mb/s stream in 1500byte pa kets, loss rate of 0.02% represents one lost pa kets in 6 se onds. These single pa ket
losses were observable as image pixelization when looking arefully at the video sequen e. We
used a rather dynami demonstrational lip from Optibase. It is likely that less dynami
sequen es, su h as broad asting from a onferen e would suer less from lost data. The ee t
of loss rate of 0.02% and 0.1% is illustrated in Fig. 2. The gure shows some of the worst
pixelizations that have o urred, these faults were interspersed with period of a eptable quality.
On the other hand, the video transmission proved to be highly resilient to delay and jitter.
We tried various ombinations of mean delay and standard deviation resulting in observations
summarized in Table 2. The system was able to adapt to the mean delay of up to 10 se onds
whi h is well beyond onditions in real-world networks.
2

Loss rate
0.01%
0.02%
0.1%
0.2%

Ee t
No observable ee t
Little pixelization in fast sequen es
Lot of pixelization and interruptions
Still image

Table 1: In uen e of pa ket loss rate on image quality

Mean delay
100 ms
100 ms
100 ms
100 ms

Standard deviation
5 ms
7.5 ms
10 ms
20 ms

Ee t
No observable ee t
No observable ee t
A short interruption of video every about 5 s
Frequent interruptions of video and audio

Table 2: In uen e of delay and jitter on image quality

Figure 2: Ee t of loss rate of 0.02% (left) and 0.1% (right)

0.06

0.14

0.04

0.135

0.02

IPDV [s]

Delay [s]

0.145

0.13

0.125

-0.02

0.12

-0.04

0.115

-0.06
0

10

Time [s]

10

Time [s]

Figure 3: Measured maximum delay (left) and jitter (right) over intervals of 100 ms produ ed
by Nist Net ongured for mean delay of 100 ms and standard deviation of 10 ms
A parti ularly interesting point is the value of standard deviation whi h was still a eptable.
For example, the mean delay of 100 ms with standard deviation of 10 ms means approximately
47% (!) of reordered pa kets. This was made possible probably be ause of layered nature of
MPEG data whi h is designed to work with ertain reordering of data pa kets [4. Delay and
jitter experien ed by the video streams in this parti ular example is illustrated in Fig. 3. These
diagrams were obtained using our system for pre ise measurement of network QoS hara teristi s [2, 3.
The video transmission was also tolerable to high pa ket dupli ations (tested up to 10%),
whi h ould be expe ted and is well beyond onditions in real-world networks.
In addition to emulation, we tried transmission over a real network between two buildings in
the university ampus. There were three routers inter onne ted with Fast and Gigabit Ethernet
links along the path. The image was free of interruptions or visible pixelization, but with slight
onstant uttering.
We did not have an opportunity to try transmission over a wirelless link, but loss rate
measurement that we performed on one of our wirelless links suggest that it may be di ult to
transmit MPEG video over a wirelless link. The pa ket loss rate measured on the Mi ro om
wirelless 10 Mb/s link from Prague to Podebrady over a period of 5 days is indi ated in Fig. 4.
There was an almost onstant loss rate of about 1.7% although the link was only lightly loaded.
Su h loss rate would probably render MPEG video transmission impossible. We urrently do
not know whether this is a standard operation ondition of the wireless te hnology used or if
there is some te hni al problem.

5 Con lusion
We found that MPEG video transmission without FEC is highly sus eptible to pa ket losses.
Maximum a eptable pa ket loss rate for awless image was approximately 0.01%. Su h pa ket
loss rate is probably di ult to a hieve on wireless links whi h means that MPEG video transmission without FEC is probably not possible over wireless links. Tests on real wireless links
would have to be ondu ted to nd out the a tual behaviour of this lass of appli ation on
4

Figure 4: Measured pa ket loss rate on the wireless link from Prague to Podebrady
wirelless links.
We also suppose that the observable image quality an depend on the de oder implementation. If the de oder ould render dupli ate frames instead of damaged frames, it would be
probably onsidered mu h less disturbing by users.

Referen es
[1 \NIST Net", Internetworking Te hnology Group (ITG), National Institute of Standards
and Te hnology (NIST), http://snad.n sl.nist.gov/itg/nistnet.
[2 Sven Ubik, Vladimir Smotla ha (Cesnet), Sampo Saaristo (Tampere University of Te hnology), Juha Laine (Soon Communi ations). \Low-Cost Pre ise QoS Measurement Tool",
Cesnet Te hni al Report 7/2001, http://www. esnet. z/do /te hzpravy/2001/07.
[3 Sven Ubik. \Presne a jednodu he meren kvalitativn h parametr
u ste" (in Cze h),
Sdelova  te hnika 2002/5.
[4 Komura Takaaki, Fujikawa Kenji, Ikeda Katsuo. \Layered Transmission and Control of
Pa ket Transmission Order for Multimedia Broad asting", Pro eedings of INET 2000,
Tokyo, 18-21.7.2000.