1
State of art of IP telephony and MPLS
L. Subirats, Telematics Engineering Dept., Technical University of Catalonia
Abstract — There are several quality of service (QoS)
requirements in order to provide a high – quality voice over IP
(VoIP) call. Initially, QoS schemes where defined for the TCP / IP
networks with Integrated Service (IntServ) architecture and then
as Differentiated Services (DiffServ) architecture. However, the
TCP/IP networks operate on the hop – by – hop basis and cannot
provide a comprehensive Traffic Engineering (TE) as
Multiprotocol Label Switching (MPLS) does. Although IP/MPLS
technology will be at the core of multiservice networks and is the
future, MPLS is still maturing. QoS standards now associated to
ATM and traffic dimensioning of Telephony over MPLS
(ToMPLS), must be improved; so there is research to do in this
field. Since VoIP is the solution available and is most likely to be
implemented in the evolution from ATM to MPLS, it would be
natural on a later stage to evolve from VoIP to VoMPLS in the
core network.
Index Terms—Telephony; MPLS; VoIP; QoS; SLA
I. INTRODUCTION
T
HERE is a plethora of published papers describing the
factors involved in a high – quality VoIP call. These
factors include the speech codec, packetization, packet loss,
delay, delay variation, and the network architecture to provide
quality of service (QoS). Other factors involved in making a
successful voice over IP (VoIP) call include the call setup
signaling protocol, call admission control, security concerns
and the ability to traverse network address translation (NAT)
and firewall. Although VoIP involves the transmission of
digitized voice in packets, the telephone itself may be analog
or digital. The voice may be digitized and encoded either
before or concurrently with packetization. Fig. 1 shows a
business in which a Private Branch Exchange (PBX) is
connected to VoIP gateways as well as to the local telephone
company central office [4].
Fig. 1. Business use of VoIP
Initially, QoS schemes were defined for the TCP / IP
networks, first as the Integrated Services (IntServ) architecture
with the Resource reservation protocol (RSVP) signaling, and
then as the Differentiated Services (DiffServ) architecture.
However, the TCP/IP networks operate on the hop – by – hop
basis and cannot provide a comprehensive Traffic Engineering
(TE). The Multi Protocol Label Switching (MPLS) technology
has emerged as a connection – oriented layer serving
connectionless IP networks, and thus it provided the means for
TE of the IP traffic. Furthermore, by implementing QoS
mechanisms in the MPLS networks, signaling paths (SPs) can
use both necessary technologies (TE and QoS) for offering
QoS guarantees to their customers.
The end – to end (E2E) path of an application flow usually
traverses several environments including the end user's
workstation and local, access and core networks. The user's
perception of quality is based on the end – to – end
performance, and QoS mechanisms of various components
must interwork for achieving the E2E QoS targets.
In conclusion, MPLS is being adopted as a key protocol
within ISP’s and carrier networks because of:
 Traffic Engineering
 Flexible signaling plane which support service
differentiation
 Flexible path definition (including redundant paths
for security functions)
 Ability to transport almost anything (i.e. voice over
IP or directly over MPLS)
 QoS techniques implemented with Diffserv or
IntServ.
The rest of the paper is organized as follows. In Sec. II we
provide an overall view of the telephony over MPLS state of
art. In Sec. III some topics such as QoS, Service Level
Agreement (SLA) and service classes of the Diff-serv/MPLS
architecture are analyzed. In Sec. IV two hot topics in this
field like “Dimensioning Aggregated Voice Traffic in MPLS
nodes” and “An End – to – End QoS Architecture with the
MPLS – Based core” are summarized in a systematic way
focusing on the problem, the steps followed to solve it, and the
final result of the research. The contribution of this paper is in
Sec V, where there is a ranking of telephony over MPLS
(ToMPLS) papers according to several variables such as the
interest, the impact and the newness of the field of the paper.
Furthermore, positive and negative opinions of references are
given, ranking them depending on the interest, impact or
newness of the field and other personal comments. Finally, in
Sec. VI, we arrive to some conclusions of the advantages and
disadvantages of telephony over MPLS comparing to other
transport technologies.
2
II. TELEPHONY OVER MPLS
We can have several voice trunking scenarios in MPLS as it
can be seen in Fig. 2. In that figure we have some examples of
Voice Networks which describe the possibilities how voice can
be transported. We can also see the MPLS network with its
gateway devices, the Label Switched Path and the Label
Switching Routers inside the network that allow providing
traffic engineering. After the MPLS network, the voice
continues being transported through another bunch of
networks.
Fig. 4. Dimensioning example for the G.729 code, Dmax=50ms and different
N values.
Fig. 2. Voice trunking transport scenarios in MPLS
Basically, telephony over MPLS can refer to two different
solutions:
 VoIP over MPLS, which consists of a number of
VoIP streams integrated in a common LSP. If we
see the stack of protocols of Fig 2. voice would be
over IP in the protocol stack.
 Voice over MPLS (VoMPLS), where a number of
voice calls may be transported over an LSP. The
multiplexing structure consists of a mandatory
outer label, zero or more inner labels, and one or
more VoMPLS primary subframes consisting of a
4-octet header (HDR) and variable length primary
payload (i.e. voice frames or a single SID frame),
as shown. In Fig 2. we can also appreciate the stack
of protocols of VoMPLS.
It can be seen in Fig 3. that VoMPLS is more efficient in
terms of bandwidth than VoIP over MPLS shown by Fig.4.
Within the header of a primary subframe, the length field is
indicated in multiples of 4 octets. Thus, up to three padding
octets can be inserted in each subframe depending on the
codec’s frame size and the number of codec’s frames carried
in the primary subframe. However, it must be considered that
other combinations of protocol stacks are possible, and they
are shown at Fig. 5.
Fig. 3. G.729 codec over MPLS, Dmax=50ms and different N values.
Fig. 5. Protocol stacks of voice with different transport technologies
3
III. QOS
A SLA is a formal contract of the relationship that exists
between a service provider and its customers. Today, people
on Internet can be divided into three roles: users, application
providers and network service providers. Users are concerned
about application price which they have to pay and user – level
QoS which they may receive, e.g., call blocking rate and voice
quality in VoIP services. Application providers care about the
balance of revenue and investment, and how to provide a
wonderful application service to users. Similarly, network
providers are concerned about the balance of revenue and
investment, and how to provide a good transport service.
How to satisfy the expectations of each role depends on a
suitable service contract and a fair and flexible QoS
management framework. In fact, although many operators who
are offering application services to the end users are often
offering Internet access service at the same time, the operators
also need a general framework of SLA management to resolve
the QoS – mapping issue and the balances of application
service and network service between revenue and cost. In Fig.
6, two VoIP users communicate which each other, and they
may or may not belong to the same VoIP provider. The VoIP
users and their VoIP service provider should contract a
vertical SLA defined as a session – level SLA (S- SLA) in this
paper, where the session is defined as a lasting relationship or
connection between a user (or user agent) and a peer on
application layer, i.e. a call. Furthermore, both VoIP users in
Fig. 6 have a horizontal relationship such as a family or a
friend.
several AS domains.
Fig. 7. QoS – oriented VoIP service over DiffServ/MPLS network
In order to commit the contract, SLA is usually contracted
loosely to tolerate the variation of system condition; however,
some application required critical QoS cannot be
compromised. Therefore, the issue of how to develop a SLA
management framework to provide a SLA predesign, optimal
QoS mapping and on – line adaptive QoS tuning is very
important. Three major concerns for such systems are
performance, availability and security. Performance
requirements imply that these systems must be adaptable and
self-configurable to the changes of workload. Availability and
security requirements suggest that these systems must also
adapt and reconfigure themselves to withstand attacks and
failures. A robust, fair and efficient SLA management model
continuously monitors the system workload and determines the
optional configuration to reach the goal of service provider in
Fig. 8. [6]. As for the Diff-Serv/MPLS architecture, three
service classes can be provided, as it can be seen in Table 1.
TABLE I
QOS – ADVANTAGES AND DISADVANTAGES OF THE THREE SERVICE CLASSES
PROVIDED FOR THE DIFF-SERV/MPLS ARCHITECTURE.
Advantages
Expedited
Forwarding (EF)
Assured
Forwarding (AF)
Best Effort (BF)
Fig. 6. SLAs between VoIP users, VoIP service provider and network provider
In reference to the system environment, the design of our
SLA management framework and QoS mapping scheme is
based on the system environments shown in Fig. 7. Unlike the
conventional best – effort VoIP service on Internet, the VoIP
service provider provides QoS – oriented VoIP service via
QoS – enabled Diffserv/MPLS networks. In addition the QoS
– enabled Diffserv/MPLS networks may be composed of
Disadvantages
VoIP end-to-end QoS
guarantee
QoS/ Cost trade off
Cost
Cost
VoIP end – to – end
QoS guarantee
QoS/Cost trade off
Regarding the table, the AF+ service model has
simultaneously the advantages of EF-class service and AFclass service. On the other hand all of these conventional
differentiated services in Diff- Serv/MPLS network only
provide one EF-class service, one AF-class service or one BFclass service to a group of VoIP users individually.
AF+ Service Model is defined as a VoIP trunk and is treat
as an EF class with α ratio and AF class with 1 − α ratio.
4
10.
According to the theorem of Birth-death system [10], the
equilibrium equations can be written as below:
Fig. 8 QoS – SLA management network
Fig 10. Two-dimension state-transition-rate diagram for the AF+ service
model
Fig 9. EF-class and AF-class provisioned bandwidth in a VoIP trunk.
Bandwidth provisioning of a VoIP trunk is contracted by a
network provider and a VoIP service provider in advance. All
of the VoIP calls in this VoIP trunk can be served as AF class
and EF class simultaneously, e.g. the gray EF-class VoIP flows
and the white AF-class VoIP flows in Fig.9. AF+ Service has
the advantages of better QoS than EF class and lower cost than
AF class in the meantime. If α approximates to 1, the network
QoS is very nice, but its cost is very expensive. Otherwise, if α
approximates to 0, the voice quality is not to be guaranteed,
but the cost is cheaper than EF class. In the below sessions, the
analysis of AF+ Service Model will be presented, and we will
give a guideline to provision network bandwidth with the
parameter α.
Under the framework of Diff-Serv/MPLS network service
model, we propose an AF+ service class for VoIP applications.
In this service, the amount of bandwidth for m EF class and n
AF class is reserved at the same time in advance. The ratio of
bandwidth provision for EF class α is shown in Eq. 1.
α = m; m + n , where 0 ≤ α ≤ 1
(1)
λP0,0 = μP0,1 + μP1,0 , if i = 0 and j = 0
(2)
(λ + iμ)Pi,0 = λPi−1,0 + (i + 1)μPi+1,0 + μPi,1 , if 1 ≤ i ≤ m −
1 and j = 0
(λ + mμ)Pm,0 = λPm−1,0 + μPm,1, if i = m and j = 0
(λ + jμ)P0,j = (j + 1)μP0,j+1 + μP1,j , if i = 0 and 1 ≤ j < ∞ (3)
(λ + iμ + jμ)Pi,j = (i + 1)μPi+1,j + (j + 1)μ P(i, j + 1) + λPi−1,j
, if 1 ≤ i ≤ m − 1 and 1 ≤ j < ∞
(λ + mμ + jμ)Pm,j = λPm,j−1 + λPm−1,j+(j + 1)μPm,j+1 , if i
= m and 1 ≤ j < ∞
Because of the characteristic of the Birth-death Process, the
constraint equations are shown in Eq. 4, Eq. 5 and Eq. 6. The
summation of all steady-states probabilities equals to 1. Eq. 5
and Eq. 6 are the probability characteristic.

m
 Pi, j  1
j 0 i 0
Pi,j = 1
Pi,j ≥ 0 , where 0 ≤ i ≤ m and 0 ≤ j < ∞
Pi,j ≤ 1 , where 0 ≤ i ≤ m and 0 ≤ j < ∞
n
(4)
(5)
(6)
m
U EF ( m,n )   iPi , j
j 0 i 0
m
When a new call arrives and the provisioned bandwidth of
EF class is available, this call is served with the EF-class level
first. Otherwise, if the provisioning bandwidth of EF class is
unavailable, it is serviced with the AF-class level and is
regarded as an AF class in the conventional Diff-Serv/MPLS
model. The proposed AF+ service are modeled as a two
dimension Birth-death Process [10], which is illustrated in Fig.
(3)
n
U AF ( m,n )   jPi , j
(7)
(8)
i 0 j 0
m
n
U cb ( m,n )  1   Pi , j
(9)
i 0 j 0
Because this two-dimension Birth-death Process is an
asymmetric system, the closed-form equation of the steady-
5
state probabilities cannot be proved directly. However, it can
be resolve by approximate analysis (LINGO) from Eq. 4, Eq. 5
and Eq. 6. If m is the number of channels and equal to 25
reserved for EF class, and the AF class bandwidth is always
available. Thus, the provisioned bandwidth utilization of EF
class and AF class are calculated from Eq. 7 and Eq. 8. and
illustrated in Fig. 11. It can be seen that most of the VoIP calls
in the VoIP trunk can be served as EF class, and their QoS can
be guaranteed well. Then a few burst arrival call can be served
as AF class.
class and AF class simultaneously. The QoS of most VoIP
calls can be guaranteed and the cost can be minimized.
IV. HOT TOPICS
Having contextualized the scenarios of telephony over
TABLE II
DIMENSIONING AGGREGATED VOICE TRAFFIC IN MPLS NODES [1]
MPLS in the previous sections, two hot topics about the
subject are explained in tables I and II together with the
solutions that researchers of the field have arrived.
Problem
Steps
Classical dimensioning models based on on-off sources
multiplexing are not accurate since they assume that no traffic is
generated during off periods. What are bandwidth requirements
for voice traffic aggregated of traffic engineering?
Step 1: Define a model for a generic voice source which
embodies any current type of voice source (which we name
Generalized VoIP source)
Step 2: Extend the fluid model with our new GVoIP source to
obtain a loss and delay prediction for each multiplexer node.
Step 3: Design a simple but efficient dimensioning algorithm
that provides the bandwidth requirement for a desired
performance when multiplexing a number of homogeneous
GVoIP sources.
Result
Using the dimensioning algorithm we compare the bandwidth
requirements for VoIP over MPLS and VoMPLS. The
estimation of the bandwidth requirement of a voice traffic
trunk over MPLS was improved.
Fig 11. The provisioned bandwidth utilization while m=25.
Now, we assume that the traffic intensity of VoIP calls
equals to 25 Erlang. The call blocking rate with the EF – class
provisioned bandwidth and AF – class provisioned bandwidth
can be calculated from Eq. 9, and the call blocking rate with
the EF – class and AF – class bandwidth provisioning as
illustrated in Fig. 12.
TABLE III
AN END – TO – END QOS ARCHITECTURE WITH THE MPLS – BASED CORE [2]
Some MPLS mechanisms are necessary in order to allow TE +
Problem
QoS guarantees.
Steps
Step 1: LAN QoS: The 802.1D-compatible switches provide
strict priority queuing and assure QoS for VoIP and other high
priority traffic.
Step 2: Core QoS:
-
DiffServ: based on a set of enhancements to the IP
protocol
which
enables
scalable
service
discrimination in an IP network. Scheduling and
packet discard are determined based on the value of
the DiffServ Code Point (DSCP) marked in the
header of an incoming IP packet.
-
MPLS support of DiffServ: The mapping between the
DiffServ BAs and the MPLS LSPs achieve the best
match for the DiffServ, TE and protection objectives
within a particular network.
Step 3: LAN-to-CORE QoS internetworking: DiffServ-toMPLS InterWorking and MPLS UNI.
Result
Fig 12. The call blocking rate with the EF – class and AF – class bandwidth
provisioning while Erlang = 25.
Therefore, the AF+ service model has the advantages of EF
Network technologies contributing to the end-to-end QoS,
including those in the customer premises networks, in the
service provider’s core, and the interworking between the LAN
and service provider core mechanisms.
V. PERSONAL OPINION OF THE REFERENCES
A ranking was elaborated taking as variables the interest of
6
networks technology.
the topic, the impact of the research and the newness of the
field. A total mark with the mean of these three variables
resulted and can be seen in Fig. 13. considering the following
references:
 [1] Dimensioning Aggregated Voice Traffic in
MPLS Nodes.
 [2] An end-to-end QoS architecture with the MPLSbased core.
 [3] Voice over MPLS compared to voice over other
packet transport technologies.
 [4] Voice over Internet protocol (VoIP).
 [5] A novel AF+ service for VoIP applications over
a Diff-Serv/MPLS network.
 [6] Session-Level and Network-Level SLA
Structures and VoIP Service Policy over DiffServBased MPLS Networks.
[4]
[5]
[6]
High but not as high as the
other parameters. / High as it
defines the basic concepts of
VoIP.
Not very recent
because it uses
studies done by
others.
Technical.
High as the numerical results
are useful for VoIP users and
network manager to contract
SLA and plan network
resource distribution.
Considerable
new as it
provides a
service that has
a compromise
between
efficiency and
price having the
benefits of the
two service
levels models.
Very high.
Not extremely
new as it
describes the
SLA structure.
Technical.
High as it implements a
framework session – level to
provide QoS- oriented
application services.
TABLE V
PERSONAL OPINION OF THE ARTICLES
Paper
Positive / Negative
[1]
(+) Reads well. (+) Good contextualization and theoretical
analysis. (+) Algorithm which leads to numerical results (+)
Good outcome. (+) Easy application to real life.
(-) Simulations made only for m=25 (m=number of channels).
(-) No proof in a real scenario.
Fig 13. Personal opinion of the articles described in the document.
[2]
The reasons for these punctuations are described at Table
IV and some personal opinion of the articles is given at Table
V.
(-) No simulation to see the improvement when using this
architecture. (-) No proof in a real scenario.
TABLE IV
REASONS OF THE ARTICLE PUNCTUATIONS
Newness of the
field
Paper
Interest / Impact
[1]
High as QoS estimation is
really important in
dimensioning current MPLS.
/ Considerable because it
describes concrete models of
aggregated traffic.
Not really
recent.
Ordinary. / It is normal as it
described various network
technologies and made
emphasis on the MPLS
mechanisms that allow to
traffic engineer the core
networks.
Ordinary, this
field has already
been
investigated.
High as it compares MPLS
with the other packet
transport technologies and it
analyzes which of them is
more efficient. / High
because it shows the viability
of a migration in the
High because it
describes the
VoIP protocol
(the article is of
a senior
member of
IEEE).
[2]
[3]
(+) Reads well. (+) Well analyzed architecture in terms of a
state of the art. (+) Advantages and disadvantages of using
each technology for each specific scenario.
[3]
Technical
(+) Summary of the state of Voice over MPLS compared to
other packet transport technologies. (+) Analysis of voice
codecs in different scenarios. (+) Talks about implementation
to real life.
(-) Neither theoretical analysis nor analysis. (-)
Neither examples nor case studies.
Very
technical.
[4]
(+) Reads well and full of examples. (+) Gives a clear
overview of VoIP.
(-) No implementation of the protocol.
[5]
Not much
technical
(-) Would be interesting to look for the bandwidth and call
blocking rate for different number of channels. (-) Scheduling
explanation is not really clear.
[6]
Not very
technical.
(+) Reads well. (+) Really positive outcome. (+) Interesting
theoretical results proved by the simulation results. (+) Nice to
do a real implementation.
(+) Reads well. (+) Several algorithms and a numeric example
which helps the reader to see the benefits of this structures.
(+) Application to different fields such as real – time
multimedia and non – real time data services.
(-) No real implementation.
7
ACKNOWLEDGMENT
This work was fully supported by “la Caixa” Foundation
through “la Caixa” Fellowship for Spanish master students.
VI. CONCLUSIONS
Several conclusions are extracted from all these analyzed
papers mentioned in the document:
 Most enterprises still use Frame Relay
 Network growth will be focused on existing
revenue-generating services and interoperability
with other existing network infrastructures in place
is needed
 MPLS is still maturing (finalizing QoS standards
now associated to ATM and traffic dimensioning
of ToMPLS)
 IP/MPLS technology will be at the core of
multiservice networks and is the future
 VoIP is the solution available, as MPLS matures,
VoIP will evolve to VoMPLS
Since VoIP is the solution available and is most likely to be
implemented in the evolution from ATM to MPLS, it would be
natural on a later stage to evolve from VoIP to VoMPLS in the
core network. A comparison between MPLS and other
transport networks can be seen at Table VI.
TABLE VI
A COMPARISON BETWEEN MPLS AND OTHER NETWORKS THAT CARRY VOICE
VoMPLS
VoIP
VoATM
VoFR
Quantitat
ive
guarantee
s.
Not
standardi
zed.
None.
QoS
Comprehensi
ve QoS on
aggregated
traffic.
Scalable with
qualitative
guarantees or
non – scalable
with
quantitative
guarantees.
Signaling
CR – LDP
and RSVP TE
Non – scalable.
PNNI.
Access
network
Software
upgrade to
customer
IAD.
Integration with
CPE IP
applications.
Standard for
cable networks.
Backbone
network
Traffic
engineering
Interworking
with
IP/DiffServ.
Implementabl
e over ATM.
No quantitative
scalable QoS.
Standard
for DSL
loop
emulatio
n.
Large
installed
base.
Proven
network
managem
ent.
Very high
with PPP.
Requires RTP
multiplexing.
Header
compression
improves
efficiency in the
access network.
Bandwidth
efficiency
Non
monotoni
c with
increasin
g traffic.
Widespre
ad
deployme
nt.
N/A.
Very
high.
VII. REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
A. Estepa, R. Estepa, I. Campos, A. Delgado. Dimensioning Aggregated
Voice Traffic in MPLS Nodes. University of Seville Seville, Spain.
41092, International Conference on Optical Network Design and
Modeling, 2008. ONDM 2008.
V. Fineberg, C. Cheng; X. XiPeng. An end-to-end QoS architecture
with the MPLS-based core. Workshop on IP Operations and
Management, 2002 IEEE Volume , Issue , 2002 Page(s): 26 – 30.
D. Wright. Voice over MPLS compared to voice over other packet
transport
technologies.
Communications
Magazine,
IEEE
Publication Date: Nov 2002 Volume: 40, Issue: 11 On page(s): 124132.
B. Goode. Voice over Internet protocol (VoIP). Proceedings of the IEEE
Volume 90, Issue 9, Sep 2002 Page(s): 1495 – 1517.
H. Su; H. Chen; C. Wang; K. Chen. A novel AF+ service for VoIP
applications over a Diff-Serv/MPLS network. Vehicular Technology
Conference, 2004. VTC2004-Fall. 2004 IEEE 60th Volume 7, Issue ,
26-29 Sept. 2004 Page(s): 4846 - 4850 Vol. 7.
H. Sui, Z. Yaui, C. Wui and K. Cheni. Session-Level and NetworkLevel SLA Structures and VoIP Service Policy over DiffServ-Based
MPLS Networks. IEICE Transactions on Communications 2006 E89B(2):383-392.
Laia Subirats received the B.S and M.S degrees in Pompeu Fabra
University, in Barcelona, Spain in 2006 and 2008, respectively. Previously,
she was an intern in Telefonica R&D in Barcelona from 2006 to 2008 and
during the summer in European Organization for Nuclear Research, Geneva,
Switzerland and participated in the program Google Summer of Code. She is
currently a telematics master student in Information Security Group in the
Department of Telematics of the Technical University of Catalonia, also in
Barcelona.