Wireless Sensor Network, 2009, 3, 132-141
doi:10.4236/wsn.2009.13019 ctober 2009 (http://www.SciRP.org/journal/wsn/).
Published Online O
A Caching Scheme for Session Setup in IMS Network
Yufei CAO1,2, Jianxin LIAO1,2, Qi QI1,2, Xiaomin ZHU1,2
1State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications,
Beijing, China
2EBUPT Information Technology Co., Ltd, Beijing, China
Received April 29, 2009; revised May 8, 2009; accepted May 10, 2009
Abstract
In IP Multimedia Subsystem (IMS), the session setup delay is a critical value for Quality of Service (QoS).
The existing approaches to improve this metric depend on optimization of Session Initiation Protocol (SIP)
message transmitting and signaling flows. Unfortunately, some service features are missing considered al-
though they have been used widely in traditional 2G networks. This paper proposes a novel session setup
scheme based on caching, upon the investigation of the performance of IMS session establishment. This
mechanism involves cache based local routing policy and an adaptive caching algorithm, which can decrease
call setup delay effectively as cached information in the terminating Serving-Call Session Control Function
(S-CSCF) hit. The analytical model is deduced, as well as the delay and cost ratio functions are presented
based on the model. Moreover, the analytical model is validated through the performance simulation in
which the performance of the proposed novel method is evaluated against the basic session setup mechanism
in terms of cost and delay.
Keywords: SIP Session, Caching, IMS
1. Introduction
IMS (IP Multimedia Subsystem) network is introduced in
3GPP R5, which aims to provide mobile user multimedia
services such as voice, video and data. It unifies core
network as all-IP network architecture and realizes the
integration of fixed and mobile communication networks
[1–3]. Home service control is selected in IMS, which
means the entity that accesses to the subscriber database
and interacts directly with service platforms is always
located at the user’s home network. Thus, location man-
agement and session management are pointed to the home
network as far as possible. The HSS (Home Subscriber
Server) contains all the information related to the users
and their services. The S-CSCF (Serving-Call Session
Control Function) located in home network provides ses-
sion control and registration services [2,4,5].
In IMS network, when the caller A wants to establish
session with the callee B, SIP INVITE request construc-
ted by UE (User Equipment) is forwarded to the user A’s
home network via the P-CSCF (Proxy-Call Session Con-
trol Function). And then user A’s home S-CSCF executes
the service control, including interaction with the AS (ap-
plication server), a process of querying DNS to determine
the entry of UE B’s home network and assigning the
S-CSCF through the I-CSCF (Interrogating-Call Session
Control Function) which is needed to select the S-CSCF
of UE B. This S-CSCF is responsible for dealing with and
ending the session, containing the interaction with AS,
sending request messages to the P-CSCF which UE B
accessed to and forwarding to UE B finally. The response
generated by UE B reverses the same path back to UE A.
After several forward and back flows, session establish-
ment is completed [6,7].
The excessive signaling of current IMS session setup
mechanism results in the long delay from the session ini-
tiation of caller and the final response of callee [8,9]. This
is unfavorable to those applications which require fast
communication handshake. Comparing with traditional
2G network, there are some problems in the flows defined
in current specifications: 1) There is no considering that
users’ session setup takes place within one S-CSCF serv-
ing area, so DNS querying and S-CSCF assignment are
involved. Thus, this process brings two problems. Firstly,
it increases unnecessary interactive signaling and the ses-
sion setup delay is raised. Secondly, the larger the scale of

Y. F. CAO ET AL. 133
users is, the more the traffic load of the entities as DNS,
I-CSCF, HSS, etc. is. That not only would be a waste of
network bandwidth and resources, but also reduces sys-
tem reliability. And eventually it leads to long session
setup delay for users’ experience. 2) A series of interac-
tive inquiries to establish sessions is too complicated.
Even if the requested destination server addresses
(S-CSCF address) of the current session request is same
as the previous one, DNS inquiry process and the
S-CSCF selection are still necessary. So it is feasible to
optimize signaling traffic load by decreasing the number
of signaling interaction.
This paper focuses on the study of basic session setup
mechanism in current IMS core network, which includes
the process of caller’s session request arriving at its
S-CSCF (the originating S-CSCF) and the connection
setup between the originating S-CSCF and the terminat-
ing S-CSCF. A cache based session setup mechanism is
proposed by improving the originating S-CSCF session
setup procedure with taking locality and caching into ac-
count. The advantages are: 1) when session participants
are in the one S-CSCF, their sessions can be established
directly. 2) If the destination server address of current
session request is the same as the previous one, session
can be established directly. It decreases the times of sig-
naling interaction for DNS query and reduces the load of
the HSS by improving the S-CSCF assignment process. 3)
There is no change of IMS core network, no adding or
varying to the terminal signaling, and no impact on sig-
naling flows. The simulation shows that the cache based
method is able to reduce signaling traffic load and session
setup delay in IMS. At the same time, as the improvement
is mainly about the signaling flows, instead of system
hardware or network structure, the cost of its implements
is smaller comparatively.
The rest of the paper is organized as follows. Section 2
reviews related work and motivates the cache based ses-
sion setup in IMS. Section 3 presents the basic session
setup flows and analyzes the terminating S-CSCF routing
process with explanation of its issues. Section 4 intro-
duces the cache based session setup mechanism with
original contribution presented in detail. Section 5 shows
the detailed analysis of the cost and the mean delay func-
tion of new mechanism. Section 6 shows evaluation of
the performance by simulation. Section 7 concludes this
paper.
2. Related Work
Usually, in Circuit Switch (CS) of both traditional 2G
and 3G networks, there are two approaches to improve
call setup performance [10].
One is to improve location management policy and
management protocols [11–14], which aims at exploring
how to manage and query user’s location information
efficiently, in order to quickly address entries serving the
callee in the core network. For example, there are
three-tier location management in [11], and layered man-
agement of mobile IP in [12,13] which restricts UE regis-
ter signaling in its local networks. Also, in [14], a cache
scheme of location information is proposed to reduce call
setup delay. In [3], two functional entities MMS (mobile
management server) and ACS (access control server) are
introduced to enhance 3G core network architecture. It
separates registration procedure and security service away
from CSCFs to achieve the efficiency of session setup by
allaying complexity of CSCFs without change to the cur-
rent IMS session setup procedure.
The other one is to improve call setup process by tak-
ing advantage of the users’ locality to advance call setup
performance [10,15]. The local routing is proposed in [10]
to modify the call setup process when the caller and callee
are in one VLR (Visitor Location Register), so the cost
between the originating MSC (Mobile Switch Center) and
the terminating HLR (Home Location Register) can be
saved. The work in [15] uses a local routing policy for
call setup based on three-tier database architecture in 3G
network with the caching in GLR (Gateway Location
Register).
Furthermore, the researches in IMS try to improve ses-
sion setup delay in two aspects. The first is optimizing
Session Initiation Protocol (SIP) signaling transmitting.
IMS session setup delay is affected by the quality of the
wireless link, e.g. frame error rate (FER), which can re-
sult in retransmissions of lost packets and can lengthen
the session setup time. One way to do is choosing the
appropriate retransmission timer and the underlying pro-
tocols. The work in [16] focuses on SIP singling transmit-
ting by optimizing it with an adaptive retransmission
timer and evaluates SIP session setup performances with
various underlying protocols, such as transport control
protocol (TCP), user datagram protocol (UDP), and radio
link protocols (RLPs). The work in [17] proposes that
choosing an appropriate SIP compression efficiency and
transport protocols can improve session setup delay. The
work in [18] studies the SIP signaling transmitting, proc-
essing and queuing delay in 3G and WiMax networks,
and proposes increasing channel rates can reduce IMS
session setup delay.
The second is improving SIP signaling flows. The
work in [8] investigates the call control procedure in
UMTS Packet Switch (PS), and decreases call setup de-
lay through performing Radio Access Network (RAN)
resource allocation concurrent with media negotiation.
But signaling interactions are reduced at the high cost of
air interface resource to achieve fast handshake. The
work in [19] is concerned about the in-calling setup de-
lay and enhances the I-CSCF reliability through check
point mechanism; also it uses the cache in I-CSCF to
accelerate session setup. To the problem of triangular
routing for a certain period of time when the user is
C opyright © 2009 SciRes. WSN

Y. F. CAO ET AL. 134
moving, Alam, M. T. et al. [20] proposes a decision al-
gorithm to select the optimal session setup option. 3. Session Setup in IMS
The features of call service in IMS are similar to those
in traditional 2G and 3G network, such as characteristics
of localization. However, the previous works seldom
take advantage of the locality to improve IMS session
setup. Therefore, we propose the cache based session
setup mechanism involving the local routing policy along
with an adaptive caching algorithm, in order to solve the
problem of signaling waste and reduce session setup delay.
3.1. Basic Session Setup Flow
Figure 1 shows a classic session setup procedure in IMS
network. With no lost of general we could suppose that
as follows:
1) UE A and UE B are IMS terminals with the same
type of properties.
Figure 1. Session establishment procedure in IMS.

Y. F. CAO ET AL. 135
2) For simple consideration, both the caller and callee
have no service contacts with the session.
3) P-CSCF # 1 and S-CSCF # 1 in the originating net-
work are the entities of P-CSCF and S-CSCF providing
services to the caller. Similarly, P-CSCF # 2 and S-CSCF
# 2 in terminating network are the entities of P-CSCF
and S-CSCF providing services to the callee.
As shown in Figure 1, when originating user A wants
to establish session with terminating user B, UE A initi-
ates a call to UE B by sending SIP ‘INVITE’. As the
‘INVITE’ request arrives at S-CSCF # 1 via P-CSCF # 1,
S-CSCF # 1 controls the services and the session. This
procedure of each flow is given below: verify initial filter
criteria (iFC) ; determine the address of network entrance
the I-CSCF by querying DNS; forward ‘INVITE’ request
to I-CSCF; the I-CSCF queries HSS to obtain the address
of S-CSCF # 2, and then forwards the ‘INVITE’ request
to S-CSCF # 2; finally when S-CSCF # 2 completes veri-
fying iFC, the ‘INVITE’ request is forwarded to termi-
nating UE B via P-CSCF # 2. Originating and terminat-
ing users take Quality of Service (QoS) negotiation by
Session Description Protocol (SDP), which is carried in
SIP message. After the two sides’ QoS negotiation, re-
source reservation (3-40 steps in Figure 1) is processed.
When UE A completes resource reservation, it notifies
UE B by ‘UPDATE’ request, and UE B responses ‘200
OK’ to confirm it. Following UE B’s completion of re-
sources reservation, ‘180 Ringing’ message is sent back
to the originating UE for ringing. When the terminating
user answers, UE B sends the ‘200 OK’ which is the
response of ‘INVITE’ request. As soon as UE A receives
this response, ‘ACK’ is sent to confirm (41–67 steps in
Figure 1) that the session between originating and termi-
nating users is established.
3.2. Determine the Address of the Terminating
S-CSCF
In IMS basic session setup procedure, the originating
S-CSCF (S-CSCF#1) is the first node that tries to forward
the SIP request based on destination address which is in
the ‘Request-URI’ field carried by SIP ‘INVITE’. The
P-CSCF and the I-CSCF are not concerned the destina-
tion address, which means they don’t inspect ‘Request-
URI’ field in the SIP request. So the originating S-CSCF
is the first point parsing the destination address, that is,
according to the ‘Request-URI’ of SIP request it deter-
mines the next-hop address in the terminating network.
During this procedure, the originating S-CSCF may find
two different types of ‘Request-URI’: ‘SIP URI’ or ‘TEL
URI’. If the ‘SIP URI’ is found, a normal SIP process is
adopted and ‘INVITE’ request is forwarded to the
I-CSCF in terminating network through multi-steps DNS
querying to determine next SIP server’ s address, which
consists of transport protocol, hostname and port number
that the I-CSCF supports [21]. And if the ‘TEL URI’ is
found in ‘Request-URI’, DNS ENUM (E.164 number and
DNS) is needed to decide the right I-CSCF address in
terminating network. After receiving the ‘INVITE’ re-
quest, the I-CSCF gets the terminating S-CSCF
(S-CSCF#2) address from the HSS by Diameter LIR
(Location-Information-Request) and Diameter LIA (Lo-
cation-Information-Answer) messages and relays ‘IN-
VITE’ request. Thus, as shown in Figure 2 (A), the route
between originating and terminating network has been set
up. The detailed signalling flow presented in the specifi-
cation [6] for the origination S-CSCF towards the termi-
nating S-CSCF is summarized as follows. 1) If the analy-
sis of the destination address determined that it belongs to
a subscriber of a different operator, the request is for-
warded to a well-known entry point in the destination
operator’s network, i.e. the I-CSCF. Then the I-CSCF
queries the HSS for current location information and for-
wards the request to the S-CSCF. 2) If the analysis of the
destination address determines that it belongs to a sub-
scriber of the same operator, the S-CSCF forwards the
request to a local I-CSCF, who queries the HSS for cur-
rent location information. Then, the I-CSCF forwards the
request to the S-CSCF.
Obviously, in originating network so many DNS que-
ries and terminating S-CSCF discovery processes have
seriously impact on the session setup delay and the cost
of network transport. For IMS network which serves tens
of thousands of users, performance will be significantly
Figure 2. To find the terminating S-CSCF.

Y. F. CAO ET AL. 136
improved if we try to save a few messages for per ses-
sion. Therefore, optimizing of IMS basic session setup
mechanism is took into account. Furthermore, by ad-
dressing the terminating S-CSCF from the originating
S-CSCF fleetly we can decrease the session setup delay
and save network resources.
4. The Cache Based Session Setup Mechanism
This part describes the details of cache based session
setup mechanism. The principles of this mechanism are:
1) If caller and callee are located in one S-CSCF serving
area, their session will be established directly; 2) the
originating S-CSCF forwards the ‘INVITE’ requests
directly to the terminating S-CSCF, instead of the
I-CSCF through DNS query; 3) the originating S-CSCF
caches the terminating S-CSCF address information ob-
tained from the previous session, and set a valid time for
the cached information, which is an exponential distribu-
tion.
4.1. Cache Based Local Routing Policy
According to the principles of the cache based session
setup mechanism, the local routing policy is adopted dur-
ing session setup. The originating S-CSCF first checks
whether the callee UE has registered on the originating
S-CSCF or not. If true, session is established directly. If
originating and terminating users are not in one S-CSCF,
the originating S-CSCF checks whether the current ter-
minating S-CSCF addresses has been cached or not, then
it can determine that the next-step is directly visiting the
terminating S-CSCF or querying DNS for retrieving the
I-CSCF addresses.
The details are as follows, see Figure 2 (B):
1) When the originating S-CSCF receives the session
setup request from caller, it checks whether the callee is
within the same S-CSCF currently (the UE learns which
the S-CSCF will be serving it through the IMS registra-
tion). If originating and terminating users are in one
S-CSCF, session is established directly; else goes to 2.
2) The originating S-CSCF queries local cache to look
for whether there has been the address of the S-CSCF
serving the callee in previous sessions. If not, the S-CSCF
performs a DNS query to retrieve the I-CSCF address,
and sends ‘INVITE’ request to the terminating S-CSCF
via the I-CSCF for session establishment, otherwise goes
to 3. The originating S-CSCF caches the terminating
S-CSCF address from the first response message (e.g.
‘183’ message). 1–4 steps in Figure 2 (B).
3) For the originating S-CSCF cached the terminating
S-CSCF address information, the originating S-CSCF
sends ‘INVITE’ request to the terminating S-CSCF to
establish session, 1’ step in Figure 2(B)
4) When the terminating S-CSCF receives originating
‘INVITE’ request, re-registration process perhaps be ini-
tiated because of the terminating user’s roaming. A new
S-CSCF is selected again in terminating home network,
and the S-CSCF (the current terminating S-CSCF) is no
longer available with returning error message. The origi-
nating S-CSCF needs to re-initiate basic session setup
process. 1–4 steps in Figure 2 (B).
4.2. Adaptive Caching Algorithm
How to cache this information decides the accuracy of
address query for the S-CSCF serving the terminating
user. When the session setup request arrives, if cache
information is effective, the originating S-CSCF hit the
terminating S-CSCF with the most prefect performance.
If the cached information in terminating S-CSCF is inva-
lid or inexistence, basic session establishment procedure
is needed, then at least the performance of cache based
session setup is not worse than that of the basic one. But
if cached information is outdated, after session arriving
at a wrong location, it should re-establish in accordance
with basic procedure, which is obviously the worst.
Therefore, the effective of cached information has a great
impact on probability of hitting terminating location in
session establishment procedure.
Consequently, we design an adaptive caching algo-
rithm in terms of the mobility patterns and the locality of
call traffic rules, i.e. the probability that caller and callee
locate in the same S-CSCF serving area is large. And we
assume there is a data buffers in S-CSCF and the buffer
size can satisfy system requirement.
Let G be the set of all out-area users.
A= {x|x：the out-area users with new location
in-formation}.
B=G-A={y|y：the out-area users with comparative
mobile stabilization}.
Define vector V. // V expresses the state of out-area
user address information.
If V(x) =TRUE
x B;
Else
x A.
End If
Define vector T(x) for the caching time of the
in-formation. // T(X) expresses the period from previous
resetting to the present time.
Assume Z is the out-area user obtained in the ses-sion
setup procedure.
If z G
Set G=G {z}, B=B {z}，V(z)=TRUE，T(z)=0;
Else

Y. F. CAO ET AL. 137
Compare the address information of this procedure
and the previous cached one.
If they are same & V(z)=FALSE
Move the address information of user z from set A to
set B;
Set V(z)=TRUE, T(x)=0;
Else If they are not same & V(z)=TRUE
Move the address information of user z from set B to
set A
Set V(z)= FALSE.
End If
End If
The judgment of cached information is:
Let the cache time threshold be D.
Let out-area callee as w.
If w G
If V(w)=TRUE
If T(w) < D
Get the address information of user w from set B.
Else If T(w) > D
Address as basic session setup procedure
End If
Else If V(w)=FALSE
Session is established according to basic method;
End If
Else If w G
Session is established according to basic method.
End If
5. Analytical Model
This section deduces the cache based session setup cost
function and mean-delay function. For simplicity, we
have only analyzed the process shown in Figure 2, with-
out considering the delay, the cost of UE, P-CSCF enti-
ties etc. Our definition of parameters is shown in Table 1.
According to [14], ,,

are defined.

means the
probability of cache valid, i.e. the address of S-CSCF
serving the terminating user has been cached in the origi-
nating S-CSCF and the called UE is in the terminating
S-CSCF service area as it receives the session request.

means the probability of cache invalid: the originating
S-CSCF hasn’t cached the address information of S-
CSCF serving the callee.

means the probability of
cache miss, i.e. the address of the terminating S-CSCF
serving callee has been cached, but the callee no longer
resides the terminating S-CSCF. According to the defini-
tion, we know



1. The cost function of basic
session setup mechanism is given by
cscbasicdns si i issf
C CCCCC

 (1)
The mean delay function of basic session setup mecha-
nism is:
cscbasicdnssi i is sf
D ddddd

 (2)
The cost function of cache based session setup mecha-
nism is:
csc csc
csc
csc
(1)(()
()
()
cachingls flsss f
dns si i is sf
ssdnssiiissf
CPCP CC
CCCCC
CCCCCC






 

 )
)
(3)
The mean delay function of cache based session setup
mechanism is:
csc csc
csc
csc
(1)(()
()
()
cachingls flsss f
dns si iis sf
ssdnssiiissf
DPdP dd
dddd d
dddddd






 
 
   (4)
Table 1. Parameters definition.
Symbol Quantity Value
/
dns dns
Cd The cost / mean time delay for performing a DNS querying 15u/15t
/
s
is
Cd
i
The cost / mean time delay for transmitting message from the originating S-CSCF to the terminating
I-CSCF 10u/10t
/
ii
Cd The cost / mean time delay for one process of the S-CSCF assignment by the I-CSCF 20u/20t
/
is is
Cd

The cost / mean time delay for transmitting message from the I-CSCF to the S-CSCF in the terminating
network 5u/5t
/
s
sss
Cd

The cost / mean time delay for transmitting message from the originating S-CSCF to the terminating
S-CSCF 25u/25t
csc csc
/
s
fsf
Cd The cost / mean time delay of the S-CSCF 30u/30t
l
P The probability of caller and callee in one S-CSCF

The probability of cache hit

The probability of cache invalid

The probability of cache miss
C opyright © 2009 SciRes. WSN

Y. F. CAO ET AL. 138
Then we try to derive probabilities of cache valid, miss,
and invalid to (3), (4). Let the user residence time in
S-CSCF as which is assumed as an exponential distri-
bution with parameter
S
t
S

, and its probability density
function is:
() St
SS
f
te



 (5)
Denote by the interval between two consecutive
calls to the terminator, and the interval between the
arrival of previous call and the time as terminator move
out the S-CSCF service area.
c
t
m
t
c()
f
t and ()
m
f
t are den-
sity functions of and . We assume that the incom-
ing call is a Poisson process, and then we have
c
tm t
() ct
cc
f
te



 (6)
According to the random observer property, we have
()()St
mSS S
rt
f
tfrdre







 (7)
We assume that the resident time of cached data is an
exponential distribution with parameter h

, so
() ht
hh
f
te



 (8)
While caller initiates a session, if the address of
S-CSCF serving the callee has been cached in the origi-
nating S-CSCF and the callee is still in cached S-CSCF,
then the result is cache valid, and

is
00
[]
()()()
cm
mchc
rc mc h
tt
c
mmcchhh c m
Shc
tttt
Pt tt t
ftftftdt dt dt







 




(9)
If the originating S-CSCF has already removed the
cached information of the address of S-CSCF serving the
callee before an incoming call, the result is cache inva-
lid

. The probability

is given by
00
1[ ]1()()
ch
hc
tt
h
rchhhccc h
ch tt
Pttftft dtdt







 
 (10)
As 1



, we can get
1hS h
chS ch
 


 

 
  (11)
6. Performance Analysis
In this section, we firstly verify the validity of analytical
model by using simulation experiments, and then we use
numerical examples to investigate the performance of the
proposed cache based session setup mechanism. For cal-
culation convenience, let basic unit of cost be u, and basic
unit of time delay be t. Parameter values are shown in
Table 1. As the session setup cost of Equation (3) are the
same as the session setup delay of Equation (4) in form,
we only take the simulation experiments for the metric of
session setup cost.
6.1. Verify Analytical Results with Simulation
Results
In our simulation, there are provided the IMS network
topology consisting UEs, the P-CSCF, the I-CSCF, the
S-CSCF and the HSS. The simulation signaling flows are
the same as Figure 2. The situations of user roaming and
session initiating are simulated by generating discrete-
events, including three types: 1) call event; 2) cache up-
date; and 3) UE roaming. To investigate the impact of
various network parameters on the performance of the
new mechanism, the probability of roaming and the
probability of caller and callee in one S-CSCF service
area are varied by using different simulation configura-
tions.
Figure 3. Simulation results of session setup cost.

Y. F. CAO ET AL. 139
Table 2. Simulation and analytical results with difference of cost (=0.5).
l
P
Ccaching
λc (1/s) SMR(λc /λs)
Simulation Analytical
Error (%)
800 1 54.9999 55 0.00%
400 2 53.81361761 54.053 0.44%
266.66667 3 50.7577782 50.739 -0.04%
200 4 49.20909 48.862 -0.71%
160 5 47.78306 47.674 -0.23%
133.33333 6 47.1684 46.858 -0.66%
114.28571 7 46.3097 46.263 -0.10%
100 8 46.5301 45.811 -1.57%
88.88889 9 45.771 45.456 -0.69%
80 10 45.3963 45.169 -0.50%
Here, the S

is five time of h

, and the SMR (Ses-
sion to Mobility Ratio), expressed as cS


, is varied
from 1 to 10. Then the cost values of two session setup
mechanisms are measured and the values of 1000 sam-
ples got from the experiment in different network con-
figuration are shown in Figure 3. Moreover, Table 2
shows the session setup cost values of the cache based
session setup mechanism in both simulation experiments
and numerical results, respectively. The cost values be-
tween simulation and model have some discrepancy due
to the number of random generated discrete-events. And
the jitter of the simulation values is also depicted in the
Figure 3. If several more session events generated during
the simulation period, the cost value of simulation is
bigger than that of analytical value, and vice versa. Al-
though the values between simulation and model have
some discrepancy, as the error rates are all under 3%,
these experiments have verified that analytical model is
consistent with the simulation results.
6.2. Session Setup Cost and Delay
According to [14,15], we defined CBR (Cost Benefit Ra-
tio) that is the ratio of to and TBR (Time
Benefit Ratio) that is the ratio of to . We
calculate session setup cost value and mean delay value in
accordance with the formulas in Section 5. The conditions
and results have been depicted in the figures.
cachi ng
Cbasic C
cachi
Dng basic D
Compared result curves are given in Figure 4 upon
three different probabilities of,,

 , with ratio results
at y-axis and x-axis expressing the probability changing
of the caller and callee in one S-CSCF. If increases,
the ratio of CBR and TBR becomes smaller. According
to y-axis, along with the increasing of cache valid prob-
ability, the advantages of cache based session setup
mechanism become more obvious. When the probability
of cache miss is bigger as
l
P
=0.5

, the performance of
cache based session setup mechanism is worse than the
performance of basic session setup mechanism
(CBR/TBR>1), because of session re-establishment after
requests arriving the wrong location. However, we note
that CBR/TBR has more than 1 just upon0.18
l
P

.
Considering the locality of call traffic (the number of
local users calling local users is a large part total call
number), the probability of callers and callees not in the
same S-CSCF is smaller, so this situation will occur less.
Therefore, generally speaking, cached based session
setup mechanism outperforms the basic one.
According to the definition of SMR, the small value of
SMR means the high mobility that UE has, vice versa.
The curves varying along with the variable value of SMR
and different have been given in Figure 5. We can
see that CBR and TBR decrease as SMR increases,
which means caching contributes to decrease the session
l
P
Figure 4. Comparison of CBR/TBR with caller and callee in
one S-CSCF.
C
opyright © 2009 SciRes. WSN

Y. F. CAO ET AL. 140
Figure 5. Comparison of CBR/TBR with . 0.5
hS
t= t
Figure 6. Comparison of CBR/TBR with . 0.4
l
P=
delivery cost and mean delay if UEs have low mobility.
Figure 6 presents the curves of CBR and TBR with three
different multiple values of residence time and the
cached data . Study from y-axis, when the value of
SMR is small (high mobility), the longer the caching
time is, the greater the values of CBR and TBR are, and
as the value of SMR is higher (low mobility) and the
caching time is shorter, the values of CBR and TBR be-
come the greater.
h
t
6.3. Query Accuracy Analysis
Let 0,0, 1

 , CBR/TBR=1, then 0.333
l
P

.
In the worst situation n, i.e. cached information is out-
dated. After session arrives at wrong location, session
should be re-established in accordance with basic proce-
dure. When 0.333
l
P

, the ratio of CBR and TBR is
equal to 1 or less than 1. This shows that enlargement of
S-CSCF serving area conduces to improve the perform-
ance of session setup mechanism based on caching. We
define the query accuracy as

 , then we get
1



. From the above analysis, the per-
formance of session setup mechanism based on caching
can be improved by increasing the query accuracy

.
When caller and callee are in one S-CSCF serving area,
no information required to cache and the performance of
session setup procedure based on caching must be better
than the basic. Then we discuss the required query accu-
racy when caller makes a call to the user in other
S-CSCF serving area. For description convenience, we
name the callee in this situation as out-area user.
From the definition of

and the Equations (9) and
(10), we can derive the expression as follow:
1c
ch




   (12)
As shown in Figure 7, as SMR increasing,

in-
creases. This is the query accuracy is added along with
the value of CBR/TBR decreasing, i.e. the performance
of session setup procedure based on caching is improved.
Upon 0.5 S t
h
t

, when SMR equals 0.333, the value of

can be got as 0.957, i.e. the probability of cache hit is
large. And

decreases a little along with the SMR
increasing. Upon 5
h
tS t

, when SMR equals to 0.333,
the value of

is 0.592.

increases fast along with
the SMR rising, and the query accuracy

is as large as
0.90 when SMR is 8.33. Thus, comparing of the two
situations, to the out-area users with less mobility, the
Figure 7. Query accuracy.

Y. F. CAO ET AL. 141
session setup mechanism based on cache has advantages
of performance. The query accuracy will increase and the
performance of cache based session establishment pro-
cedure will be improved while the effective time of
cached data is prolonged.
7. Conclusions
How to decrease the network load and session delay is an
important issue for designing and deploying IMS network.
This article proposes: a cache based session setup mecha-
nism, mainly through improving signaling flows in ses-
sion establishment and introducing the caches to reduce
network load and session delay. As it mostly perfects
signaling flows, with no change of IMS core network
architecture, no change or adding to terminal signaling,
the cost of this improvement is comparatively smaller.
Therefore, this innovation is practical for building of IMS
network.
8. Acknowledgements
This work was jointly supported by: 1) National Science
Fund for Distinguished Young Scholars (No. 60525110);
2) National 973 Program (No. 2007CB307100, 2007
CB307103); 3) Development Fund Project for Electronic
and Information Industry (Mobile Service and Applica-
tion System Based on 3G).
9. References
[1] H. Y. Dong, S. Kim, C. Nam, et al., “Fixed and mobile
service convergence and reconfiguration of telecommuni-
cations value chains,” Wireless Communications, IEEE,
Vol. 11, No. 5, pp. 42–47, October 2004.
[2] 3GPP TS 23.228, V.8.2.0, IP Multimedia System (IMS),
September 2007.
[3] K. Shuang and F. C. Yang, “Study on enhanced 3G core
network architecture,” ACTA ELECTRONICA SIN ICA,
Vol. 34, No. 7, pp. 1189–1193, July 2006.
[4] 3GPP TS 22.228, V8.2.0. Service requirements for IP
multimedia core network subsystem (IMS), October
2007.
[5] 3GPP TS 23.218, V7.7.1. IP Multimedia(IM) session
handling; IM call model, September 2007.
[6] 3GPP TS 24.228, V5.15.0, Signaling flows for the IP
multimedia call control based on SIP and SDP, October
2006.
[7] 3GPP TS 24.229, V8.1.0. IP multimedia call control
protocol based on SIP and SDP, September 2007.
[8] K. Umschaden, I. Miladinovic, S. Bessler, and I. Gojmerac,
“Performance optimizations in UMTS switched call
control,” Fifth IEE International Conference on 3G
mobile communication technologies, pp. 73–177, 2004.
[9] G. Foster, M. I. Pous, D. Pesch, A. Sesmun, and V.
Kenneally, “Performance estimation of efficient UMTS
packet voice call control,” Proceedings of the IEE Fall
Vehicular Technology Conference, pp. 1447–1451, 2002.
[10] H. Jiang, B. Lu, and L. M. Li, “A novel call setup
mechanism in a GSM network,” Jounal of China Institure
of Communications, Vol. 23 No. 8, pp. 52–58, October
2002.
[11] 3GPP TS 23.119, V7.0.0, Gateway Location Register
(GLR), July 2007.
[12] X. Jiang and I. F. Akyildiz, “A novel distributed dynamic
location management scheme for minimizing signaling
costs in mobile IP,” IEEE Transactions on Mobile
Computing, Vol. 1, No. 3, pp. 163–175, July–September
2002.
[13] W. Ma and Y. Fang, “Improved distributed regional
location management scheme for mobile IP,” IEEE
Proceedings for Personal, Indoor and Mobile Radio
Communications, (PIMRC’03), Beijing, China, Vol. 3,
pp. 2505–2509, September 7–10, 2003.
[14] C. W. Pyo, J. Li, and H. Kameda, “A caching scheme for
dynamic location management in PCS network,” IEEE
58th Vehicular Technology Conference, Lbaraki, Japan,
Vol. 2, pp. 761–765, 2003.
[15] H. Zhang, J. X. Liao, and X. M. Zhu, “A novel call setup
mechanism based on three-tier databases in 3G,” Journal
of Electronics & Information Technology, Vol. 29, No. 6,
pp. 1290–1294, June 2006.
[16] H. Fathi, S. S. Chakraborty, and R. Prasad, “Optimization
of SIP session setup delay for VoIP in 3G wireless
networks,” IEEE Transactions on Mobile Computing,
Vol. 5, No. 9, pp. 1121–1132, September 2006.
[17] M. Melnyk and A. Jukan, “On signaling efficiency for
call setup in all-IP wireless networks,” IEEE International
Conference on Communications (ICC), Vol 5, pp. 1939–
1945, June 2006.
[18] A. Munir, “Analysis of SIP-based IMS session
establishment signaling for WiMax-3G networks,” Fourth
International Conference on Networking and Services
(icns), pp. 282–287, 2008.
[19] Y. B. LIN and M. H. TSAI, “Caching in I-CSCF of
UMTS IP multimedia subsystem,” IEEE Transactions on
Wireless Communications, Vol. 5, No.1, pp. 186–192,
January 2006.
[20] M. T. Alam and Z. D. Wu, “Comparison of session
establishment schemes over IMS in mobile environ-
ment,” 2005 Fifth International Conference on Informa-
tion, Communications and Signal Processing, pp. 638–
642, December 2005.
[21] J. Rosenberg and H. Schulzrinne, Session Initiation Pro-
tocol (SIP): Locating SIP Servers, RFC 3263, IETF, June
2002.