Transcript Static Call Admission Control and Dimensioning of Media Gateways

Static Call Admission Control and Dimensioning
of Media Gateways in IP based Mobile Core
Networks
Mika Isosaari
Supervisor: prof Jorma Virtamo
Instructor: Harri Lehtomäki, M.Sc.
Contents










Introduction
General Structure of UMTS Release 5 Network
Media Gateway
Multiservice IP Transport Network
Network Dimensioning
Quality of Service
Mechanisms to Guarantee QoS
Static Admission Control
Simulations
Conclutions and Future Work
© Ericsson AB 2005
2
2005-08-03
Introduction
 Background
– VoIP vs. ToIP
– How telecom grade speech can be transferred in
connectionless IP network?
– Multiservice IP network: speech only one of the services
 Objectives
– To study how circuit-switched speech can be transferred in
an IP multiservice network so that a certain Quality of
Service (QoS) level can be sustained
– How static admission control methods work and what is
their influence on network dimensioning
© Ericsson AB 2005
3
2005-08-03
Research methods
 Literature study
– ITU, 3GPP, IETF recommendation and specifications
– Books and articles to get a more comprehensive picture of
the subject
 Numerical evaluation
– Used in comparing different static admission control
methods and their effect on dimensioning
 Simulations
– Show how the traffic intensity affects the utilization and
resource demand
© Ericsson AB 2005
4
2005-08-03
General Structure of UMTS Release 5 Network
 Three domains: Circuit-Switched (CS), PacketSwitched (PS) and IP Multimedia Subsystem (IMS)
– This thesis focuses on CS domain
UTRAN
MGW
MGW
Iu
Nb
PSTN / Legacy
/ External
Mc
Mc
A/ Iu
Nc
GERAN
MSC server
GMSC server
A/ Iu
D
CAP
C
HLR
Applications
& Services
© Ericsson AB 2005
5
Signalling Interface
Signalling and Data
Transfer Interface
2005-08-03
Layered Architecture
 Application layer
 Network control layer
 Connectivity layer
– MGW
– Backbone
© Ericsson AB 2005
6
2005-08-03
Media Gateway
 PSTN/PLMN transport termination point
 May support media conversion, bearer control and
payload processing (e.g. transcoders and echo
cancellers)
 Nb User Plane traffic between MGWs is transported
either over ATM or IP bearer
 Logically resides at the border of the backbone,
physically part of site configuration
 Basic site infrastructure: Local Area Network (LAN)
switches and site routers
© Ericsson AB 2005
7
2005-08-03
Multiservice IP Transport Network
© Ericsson AB 2005
8
2005-08-03
Telephony services in multiservice IP network
 Strict requirements for Telephony over IP (ToIP)
– when international telecommunication networks interwork
with IP-based networks, the QoS experienced by the users
should, as far as practicable, be the same as if there had
been no interworking involved
 Data Conversions and Protocols
– MPLS, IP/UDP/RTP/NbUP, AMR/PCM…
© Ericsson AB 2005
9
2005-08-03
Network Dimensioning
 The whole planning process is influenced by the UMTS
architecture and IP backbone when compared to
traditional GSM network
– overall architecture is very different
– multiservice network
– information is transferred in a form of packets in a
connectionless network
 Dimensioning Challenges
– every traffic flow has an effect on all the other traffic flows
and a wrongly configured service can lead to degradation of
speech quality, which is not acceptable
– when the speech is packet-based everything comes in
practice a matter of probabilities
© Ericsson AB 2005
10
2005-08-03
Quality of Service
 Quality of Service (QoS) is the quality of a requested
service as perceived by the customer and always meant
end-to-end
 Information Quality Parameters: delay, jitter, BER, PLR,
data rate
 QoS Architecture in UMTS Networks vs. QoS in Internet
– Mapping of different quality classes important
– E.g. with DiffServ: EF  conversational, AF  streaming /
interactive, BE  background
 Internet QoS: IntServ, DiffServ, MPLS(?), traffic
engineering
© Ericsson AB 2005
11
2005-08-03
Mechanisms to Guarantee Quality of Service
 Network Level Mechanisms
– Dimensioning
– Overprovisioning
– Architecture
 Flow Level Mechanisms
– Static Admission Control
– Dynamic Admission Control
 Packet Level Mechanisms
Admission
Control
Static
Admission
Control
Pipe
Model
© Ericsson AB 2005
12
Domain
Model
Dynamic
Admission
Control
Hose
Model
MBAC
Probing
Bandwidth
Signalled
Provisioning
Broker
2005-08-03
Static Admission Control
 Basic idea: permanently allocated resources in the
backbone network set by the service provider
 MGW is in practice the most logical choice in the CN
for the implementation (may work together with routers
in the backbone)
 Main advantage of static methods is their simplicity
 Downside is the inefficient usage of network resources
 Two most important models: pipe model and hose
model
© Ericsson AB 2005
13
2005-08-03
Static Admission Control
 Pipe model
– traditional model how provisioning has been performed in
private networks
– point-to-point connection with a given pre-allocated capacity
– destination-specific: large number of configuration
parameters
– Implementation: MGW or MGW / edge router
 Hose model
– first proposed as a flexible model for resource provisioning
in VPNs
– no individual pipes between nodes but “hoses”, which
contain all incoming or outgoing traffic
– Advantages: flexibility, ease of specification, multiplexing
gain and characterization
© Ericsson AB 2005
14
2005-08-03
Simulations
 What is the actual gain of statistical multiplexing when
the traffic is handled as an aggregate rather than as
individual pipes?
 Simulations were performed with the NS2 network
simulator
– PCM: CBR UDP application
– AMR: two Exp on/off UDP apps.
PCM
AMR/speech
AMR/silence
Packet size (bytes)
91
82
56
Sending interval (ms)
5
20
160
145,6
32,8
2,8
-
600/400
400/600
90
90
90
Sending data rate (kbps)
Average on/off times (ms)
Average call duration (s)
© Ericsson AB 2005
15
N1
Access
link
Core link
R
D
N2
2005-08-03
Simulations - results
 Bandwidth
limit for link
1
0.95
– PLR 10-4
– Jitter
<5ms
0.85
0.8
Utilization / %
 Utilization:
gained link bw
divided with
average bw
0.9
0.75
0.7
0.65
0.6
0% PCM
10% PCM
33% PCM
0.55
0.5
1
10
2
10
10
3
10
Total traffic intensity / Erl
© Ericsson AB 2005
16
2005-08-03
4
Simulations - results
50 Erl
1500
34%
1000
Bandwidth / kbps
500
5000
4
x 10
1.2
5200
5400
5600
5800
6000
6200
6400
6600
6800
7000
500 Erl
13%
1.1
1
0.9
5000
5200
5400
5600
5800
5
1.08
x 10
6000
6200
6400
6600
6800
7000
6200
6400
6600
6800
7000
5000 Erl
1.06
1.04
1.02
1
5000
© Ericsson AB 2005
3%
5200
5400
5600
17
5800
6000
time / s
2005-08-03
Conclusions
 Although various mechanisms exist for guaranteeing
some QoS level in an IP network there is no particular
mechanism that alone could sustain a certain QoS 
available mechanisms should be used together so that
different mechanisms on packet, flow, and network
level complement each other
 Justifies also the use of static admission control
methods, with which a permanent limit can be set for
the traffic that a site can offer to a backbone network
© Ericsson AB 2005
18
2005-08-03
Conclusions
 Pipe vs. hose
– flexibility and easy implementation are clearly
characteristics of the hose model
– overprovisioning factor related to configuration parameters
can with high probability be kept under 2 for the hose model
– simulations show clearly that the utilization improves when
the traffic intensity is increased, but…
 Already 250 Erl traffic has utilization rate of ca. 80 %
– gain is not necessarily that significant and does not alone
make a clear difference between the two models
© Ericsson AB 2005
19
2005-08-03
Future work
 measurements from a real network are needed to
validate any simulation results
 edge router based pipe model
 dynamic resource allocation
 optimal routing method for the hose model
 domain model: combine best features from pipe and
hose models
© Ericsson AB 2005
20
2005-08-03
Questions?