Anda di halaman 1dari 4

49th International Symposium ELMAR-2007, 12-14 September 2007, Zadar, Croatia

The Impact of Increased Video Traffic on Quality of Service Parameters in


Next Generation IP/MPLS Network
Jasmina Barakovi 1, Himzo Bajri 1, Amir Husi

BH Telecom, Joint Stock Company, Sarajevo; Obala Kulina bana 8, 71 000 Sarajevo, Bosnia and Herzgovina
SIPA State Investigation and Protection Agency; Adema Bu e 102, 71 000 Sarajevo, Bosnia and Herzegovina
E-mail: jasmina.barakovic@bhtelecom.ba, himzo.bajric@bhtelecom.ba, ahusic@sipa.gov.ba
Abstract Insufficient Quality of Service (QoS) of the emerging multimedia applications is a growing concern
that has led the need for research and study. In this paper we investigate the impact of increased video traffic on
QoS parameters and their correlation in next generation network. Multi-protocol Label Switching (MPLS) and
Differentiated Services (DiffServ) integration is very useful strategy for todays traffic. MPLS Traffic
Engineering (TE) plays an important role in the implementation of network services with QoS guarantees . The
aim of simulation study is to underline how MPLS TE and MPLS Diffserv integration improve the performances
of todays networks, and identify opportunities for improvement, and development of new mechanisms to ensure
QoS features in future networks.
Keywords DiffServ, IP, MPLS, QoS
1. INTRODUCTION
The startling growth of Internet technology,
coupled with the relatively low deployment cost of
Internet Protocol (IP) networks, has created a push
for an integrated IP-based core- a single network
for data, video and voice access. The diverse service
requirements and novel traffic characteristics of the
new applications have posed many technical
challenges that must be addressed in the near future.
The paper focuses on emphasizing how Multiprotocol Label Switching (MPLS) Differentiated
Services (DiffServ) integration and MPLS Traffic
Engineering (TE) dramatically improve the traffic
performance using Network Simulator (ns-2).
Section 2 summarizes key technology components
of Quality of Service (QoS) in IP/MPLS network. It
focuses on the theoretical concepts behind MPLS
support of DiffServ architecture and presents how
MPLS TE can help address the network congestion.
Simulation model was chosen to demonstrate the
concepts introduced in the previous section. The
simulation results are briefly discussed in section 3.
Section 4 concludes the paper.
2. QUALITY OF SERVICE AND TRAFFIC
ENGINEERING IN IP/MPLS NETWORK
Two QoS architectures have been defined for IP:
IntServ (Integrated Services) and DiffServ. IntServ
provides granular QoS guarantees with explicit
resource reservation. DiffServ provides a coarse
QoS approach based on aggregates (classes) of
traffic. MPLS does not define new QoS
architectures. It provides support for DiffServ.
MPLS DiffServ introduces the concepts of EXP-

inferred-class LSP (E-LSP) and Label-inferred-class


LSP (L-LSP), along with the concept of tunnel
modes (pipe, short-pipe, and uniform) [1]. Both
architectures use a collection of traffic-control
mechanisms that include classification, marking,
policing, shaping, congestion management, active
queue management, fragmentation/interleaving, and
header compression [2].
MPLS provides native TE capabilities that can
improve network efficiency and service guarantees.
These MPLS TE capabilities bring explicit routing,
constraint-based routing, and bandwidth reservation
to MPLS networks. MPLS TE supports its routing
and bandwidth-reservation capabilities per class
through the DiffServ-Aware TE extensions and
provides a fast protection mechanism for link and
node failures using fast reroute [3].
3. SIMULATION
3.1 Simulation Aims and Environment
The aim of this simulation is to underline the
need of integration of MPLS with DiffServ in order
to improve the network performances. MPLS
rerouting is shown in this simulation as the
motivating reason behind the MPLS TE. To
investigate possible solutions for improving QoS, a
number of simulations were undertaken using ns-2.
3.2 Simulation Setup and Details
The simulation was chosen to illustrate the
improved performance brought by MPLS DiffServ
integration and MPLS TE to achieve the desirable
QoS. To demonstrate this capability, different

55

49th International Symposium ELMAR-2007, 12-14 September 2007, Zadar, Croatia

scenarios were setup to demonstrate the better


performance by implementing comparison between
IP network, DiffServ IP network, MPLS network
and DiffServ MPLS network, respectively. Four
different simulation scenarios were based on the
common topology (Fig.1). The network topology
was setup with two paths, such that default routing
would be along the shortest path. The network
consists of 11 IP nodes. In all simulation scenarios,
nodes n0, n1, n2 are source nodes, while n8, n9, n10
act as the destination nodes.
In the IP network topology, all links were set up
as duplex with 5 ms delay and use DropTail
Queuing. The duplex links between all nodes have
10 Mbps bandwidth and 5 ms delay except the link
from n3 to n5 with link capacity of 5 Mbps and with
10 ms delay. Therefore, the link from n3 to n5 will
be a bottleneck being the shortest path between
source and destination nodes.
The DiffServ IP network topology is similar to
the previous one. The difference is that there is the
simplex link from n3 to n4 which uses the dsRED/edge,
which marks the packets. Time Sliding Window with

3 Color Marking (TSW3CM) policer is used to


determine how to mark and prioritize the packet
according to user requirements. During the time of
congestion, more low priority packets are discarded
than high priority packets because Priority (PRI)
scheduling mode is used. The link from n4 and n3
uses dsRED/core which drops packets, in a manner
similar to Random Early Detection (RED), except that
video, voice and data packets are given different
treatment. The link from n5 to n4 uses dsRED/edge and
the link from n4 to n5 uses dsRED/core.

MPLS network topology is exactly the same as


the IP network topology except nodes n3, n4, n5, n6
and n7 are also MPLS capable, which allows the
non-shortest path links to be used. DiffServ MPLS
network topology uses similar network design with
the only difference being that all links between
MPLS capable nodes use Class Based Queueing
(CBQ).
To understand the behavior of different types of
traffic under various network conditions, the
performance metrics of Transmission Control
Protocol (TCP) flows as well as User Datagram
Protocol (UDP) flows were measured and compared.
TCP flows were used mostly for the shortest
transmission and normal data flow whereas UDP
was used for voice and video flows. A "TCP" agent
is attached to n0, and a connection is established to a
TCP "sink" agent attached to n8. As default, the
maximum size of a packet that a "TCP" agent can
generate is 500 Byte. A TCP "sink" agent generates
and sends ACK packets to the sender ("TCP" agent)
and digests the received packets. A "UDP" agent is
attached to n1 and to n2, which have direct
connection with n9 and n10, respectively. These
nodes are "null" agents, which just frees the packets
received.

56

Fig. 1. Simulation topology


File Transfer Protocol (FTP) traffic generator is
attached to "TCP" agent. Constant Bit Rate (CBR)
and Exponential (EXP) traffic generators are
attached to "UDP" agents. The EXP agent is
configured to generate 200 Byte voice packets with
burst time of 180 ms and idle time of 120 ms. The
CBR agent is configured to generate 1000 Byte
video packets. To analyze and to evaluate the
performance impact of mixing TCP and UDP traffic,
the simulations were run in three series with three
different sending rate of CBR traffic generator. The
sending rate of CBR traffic generator for each series
is 50%, 70% and 90% of the shortest path capacity.
The idea is to dramatically increase CBR traffic and
to analyze its impact on the behavior of FTP and
EXP traffic flows in four different network
environments. The output trace file from the
simulation is used to measure the traffic
performances (throughput, delay, jitter, loss) and
investigate the existence of correlation between
them.
3.3 Simulation Results and Analysis
The simulation results clearly show that pure IP
network only provides best effort service for both
FTP and EXP traffic flows. All traffic flows use the
shortest path (n3_n4_n5) and exceed its capacity,
while a longer path (n3_n6_n7_n5) is under-utilized.
The congestion is occurring within the network
when CBR traffic generator increases its sending
rate. Packets from link get dropped and delayed as
buffers overflow because the resources in the
network cannot meet all traffic demands.
The average throughput of both FTP flow and
EXP flow decreases in IP network environment
according to the increase of CBR traffic. Fig. 5
shows clearly that DiffServ IP network has the
advantage over the IP network in terms of
throughput of EXP flow. The reason why the
DiffServ IP network shows a substantial impact on
improving throughput of EXP flow is due to the
DiffServ-enable node using the priority queue
mechanisms. It starves the FTP traffic by allowing
them to have only a small amount of link capacity.
Consequently, the throughput of FTP flow in

49th International Symposium ELMAR-2007, 12-14 September 2007, Zadar, Croatia

DiffServ is lower than in the IP only network. Fig. 2


reflects exactly such behavior. In both MPLS and
DiffServ MPLS simulation scenarios, traffic
engineering using Constraint-based Routing Label
Distribution Protocol (CR-LDP) is applied to switch
the FTP traffic flow through an explicit route
(n3_n6_n7_n5). Fig. 2 shows that when separate
label switched path (LSPs) are used for routing
different traffic flows, FTP throughput remains
unaffected by the level of CBR traffic since each
type of traffic follows separate MPLS paths. This
configuration provides better throughput for FTP
than cases where all traffic types are allowed to mix
on a single link.
The overall performance of the network in terms
of delay is benefited by introducing DiffServ
approach. Average delay of EXP flow between
DiffServ and IP only network is shown in Fig. 6,
which demonstrates that DiffServ exhibits better
performance than IP only. Average delay of FTP
flow is shown in Fig.3. The delay curve shows that
the experience of the FTP flow in DiffServ network
is worse than in IP-only in terms of delay. The high
delay for the FTP traffic is because the low priority
queue, which becomes full quickly as packets have
to wait for the high priority queue to finish sending
premium data first. Both MPLS and DiffServ MPLS
simulations result in the reduction of the average
delay of both FTP and EXP traffic flows.
The jitter curves of both FTP and EXP traffic
flows are shown in Fig. 4 and Fig. 7. The jitter of
both FTP and EXP flows behaves the same as the
delay, since it represents variation in the delay.

Average jitter of FTP traffic according to the percentage

Average delay of FTP traffic according to the percentage

Average throughput of FTP traffic according to the percentage


increase of CBR traffic in different network environment

Table 1 demonstrates the packet loss behavior of


both FTP and EXP traffic recorded in IP, DiffServ
IP, MPLS and DiffServ MPLS network, respectively.
The FTP and EXP packets experience packet loss
only in IP network. When the first packet reaches
the bottleneck packets are forced to queue in the
buffer. The buffer queue becomes full quickly when
the CBR send packets at a specified rate. The
packets start to drop when the router simply cannot
handle the large amount of traffic, which results in
the buffer overflowing.
Table 2 and 3 summarize the correlation
coefficients between QoS parameters of different
traffic flows in IP network and DiffServ MPLS
network. These tables show how different QoS
parameters influence each other in the CBR traffic
increasing conditions. The results confirm the
existence of high positive and negative correlations
between the increased throughput of CBR flow and
other QoS parameters, especially in IP network.
High negative correlation exists between increased
throughput of CBR traffic and throughput of FTP
and EXP traffic. There is very high positive
correlation between throughput of CBR traffic and
delay, jitter and packet loss ratio (PLR) of FTP and
EXP traffic flows.
These correlation coefficients are completely
different in the DiffServ MPLS network. A number
of them do not exist because QoS mechanisms are
implemented and FTP flow uses the non-shortest
path. Therefore, the correlation between the
throughput of CBR and FTP traffic flows does not
exist.
increase of CBR traffic in different network environment

increase of CBR traffic in different network environment


0.4

0.03

IP Network

IP Network

DiffServ IP Network

DiffServ IP Network

0.35

IP MPLS Network

0.025

IP MPLS Network

2.5

DiffServ MPLS Network

eragAvothuptofFTPflw[Mbps]

DiffServ MPLS Network


0.02

eAragvydolfFTPflw[s]

aAvegrjitofFTPsflw][

0.3
2

0.25
0.015
0.2

1.5

0.01
0.15
1

0.005
0.1
IP Network

0.5

DiffServ IP Network
IP MPLS Network

0.05

DiffServ MPLS Network


0

50

40

60

70

80

-0.005

90

40

100

50

60

70

80

90

50

40

100

Fig. 3. Delay of FTP traffic flow

Fig. 2. Throughput of FTP traffic flow

Average throughput of EXP traffic according to the percentage


increase of CBR traffic in different network environment
0.05

0.3

0.044

0.294

100

increase of CBR traffic in different network environment

DiffServ IP Network

DiffServ IP Network

IP MPLS Network
DiffServ MPLS Network

DiffServ MPLS Network


eragAvjitofEXPw[sl]

0.296

eragAvdlyofEXPflw[s]

aAvehgroutp fEXPflow[Mbps]

0.046

90

IP Network

IP MPLS Network
0.298

80

Average jitter of EXP traffic according to the percentage


x 10

IP Network

0.048

70

Fig. 4. Jitter of FTP traffic flow

Average delay of EXP traffic according to the percentage


increase of CBR traffic in different network environment

0.302

60

Percent of CBR flow according to the shortest path capacity [%]

Percent of CBR flow according to the shortest path capacity [%]

Percent of CBR flow according to the shortest path capacity [%]

0.042

0.292

0.04

0.29

0.038

0.288

0.036

0.286

0.034

IP Network
DiffServ IP Network

0.284

0.032

IP MPLS Network
Diffserv MPLS Network

0.282

40

50

60

70

80

90

100

Percent of CBR flow according to the shortest path capacity [%]

Fig. 5. Throughput of EXP traffic flow

0.03

40

50

60

70

80

90

100

Percent of CBR flow according to the shortest path capacity [%]

Fig. 6. Delay of EXP traffic flow

-1
40

50

60

70

80

90

100

Percent of CBR flow according to the shortest path capacity [%]

Fig. 7. Jitter of EXP traffic flow

57

49th International Symposium ELMAR-2007, 12-14 September 2007, Zadar, Croatia

Table 1. Packet Loss of FTP and EXP traffic flows


CBR uses 50% of the shortest
path capacity
Packet Sent
Packet Lost
FTP
EXP
FTP
EXP
5643
4478
56
21
6270
4479
0
0
7691
4309
0
0
7691
4309
0
0

IP Network
DiffServ IP Network
IP MPLS Network
DiffServ MPLS Network

CBR uses 70% of the shortest


path capacity
Packet Sent
Packet Lost
FTP
EXP
FTP
EXP
3382
4478
90
81
3462
4478
0
0
7691
4309
0
0
7691
4309
0
0

CBR uses 90% of the shortest


path capacity
Packet Sent
Packet Lost
FTP
EXP
FTP
EXP
1588
4478
112
427
613
4478
0
0
7691
4309
0
0
7691
4309
0
0

FTP

EXP

CBR

PLRFT

JiterFTP

PLREX

PLRCB

JiterEXP

JiterCBR

DelayFTP

DelayEXP

DelayCBR

Througpt

Througpt

Througpt

Table 2. Correlation coefficients between QoS parameters of different traffic types in IP network

Throughput CBR
Throughput FTP
Throughput EXP
Delay CBR
Delay FTP
Delay EXP
Jitter CBR
Jitter FTP
Jitter EXP
PLR CBR
PLR FTP
PLR EXP

1.000
-0.983
-0.954
0.489
0.880
0.851
0.932
0.941
0.821
0.917
0.960
0.915

-0.983
1.000
0.884
-0.322
-0.779
-0.741
-0.849
-0.864
-0.703
-0.829
-0.893
-0.826

-0.954
0.884
1.000
-0.727
-0.982
-0.969
-0.998
-0.999
-0.954
-0.994
-0.999
-0.994

0.489
-0.322
-0.727
1.000
0.884
0.875
0.773
0.756
0.899
0.797
0.714
0.800

0.880
-0.779
-0.982
0.884
1.000
0.998
0.993
0.989
0.994
0.997
0.978
0.997

0.851
-0.741
-0.969
0.875
0.998
1.000
0.984
0.978
0.998
0.989
0.964
0.991

0.932
-0.849
-0.998
0.773
0.993
0.984
1.000
0.999
0.973
0.999
0.996
0.999

0.941
-0.864
-0.999
0.756
0.989
0.978
0.999
1.000
0.966
0.998
0.998
0.998

0.821
-0.703
-0.954
0.899
0.994
0.998
0.973
0.966
1.000
0.981
0.948
0.982

0.917
-0.829
-0.994
0.797
0.997
0.989
0.999
0.998
0.981
1.000
0.992
1.000

0.960
-0.893
-0.999
0.714
0.978
0.964
0.996
0.998
0.948
0.992
1.000
0.991

0.915
-0.826
-0.994
0.800
0.997
0.991
0.999
0.998
0.982
1.000
0.991
1.000

FTP

EXP

CBR

PLRFT

JiterFTP

PLREX

PLRCB

JiterEXP

JiterCBR

DelayFTP

DelayEXP

DelayCBR

oThugprt

oThugprt

Througpt

Table 3. Correlation coefficients between QoS parameters of different traffic types in DiffServ MPLS network

Throughput CBR
Throughput FTP
Throughput EXP
Delay CBR
Delay FTP
Delay EXP
Jitter CBR
Jitter FTP
Jitter EXP
PLR CBR
PLR FTP
PLR EXP

58

1.000
0.000
0.459
0.961
0.000
0.867
0.941
0.000
-0.964
0.000
0.000
0.000

0.000
1.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

0.459
0.000
1.000
0.197
0.000
-0.044
0.133
0.163
-0.207
0.000
0.000
0.000

0.961
0.000
0.917
1.000
0.000
0.971
0.998
0.999
-0.999
0.000
0.000
0.000

0.000
0.000
0.000
0.000
1.000
0.000
0.993
0.000
0.000
0.000
0.000
0.000

0.867
0.000
-0.044
0.971
0.000
1.000
0.984
0.000
-0.968
0.000
0.000
0.000

0.941
0.000
0.133
0.998
0.993
0.984
1.000
0.000
-0.997
0.000
0.000
0.000

0.000
0.000
0.163
0.999
0.000
0.000
0.000
1.000
0.000
0.000
0.000
0.000

-0.964
0.000
-0.207
-0.999
0.000
-0.968
-0.997
0.000
1.000
0.000
0.000
0.000

0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

4. CONCLUSION

REFERENCES

IP/MPLS is and will be a key architectural


component of voice, video and data services for next
generation networks [4]. This paper discusses and
verifies the role of QoS and TE in IP/MPLS network
through various simulations and analysis. The
overall performance of the network in terms of
throughput, delay, jitter and loss is benefited by
introducing DiffServ approach combined with
MPLS network environment. In pure IP network
exists a strong positive or negative correlation
between QoS parameters when video traffic
increases. The correlation coefficients in MPLS
DiffServ network are completely different and show
the improved traffic performances.

[1] J. Barakovi , H. Bajri , N. Daferovi , MPLS


support of DiffServ QoS Architecture, BIHTEL
2006, Sarajevo, B&H, November 2006.
[2] S. Alvarez, QoS for IP/MPLS Networks, Cisco
Press, June 2006.
[3] J .Guichard, F. Le Faucheur, J.P. Vasseur,
Definitive MPLS Designs, Cisco Press, March
2005.
[4] M. Morrow, A. Sayeed, MPLS and NextGeneration Networks: Foundations for NGN
and Enterprise Virtualization, Cisco Press,
November 2006.

Anda mungkin juga menyukai