Communication Cost of SIP Signaling in Wireless Networks and Services

Abstract

Developers of SIP mostly concentrated on making SIP to do what no other protocol can do. In a way they failed to consider the weaknesses and dangers. From this realization we did a study on the weaknesses of SIP and SIP signaling in Wireless Networks and Services. From our study we identified Delay and Security as the most severe problems, which we discuss in this paper.

Full text

Page 1

173

1
Communication Cost of SIP Signaling in
Wireless Networks and Services
Elthea T. Lakay, Johnson I. Agbinya**
Private Bag X17, University of the Western Cape, Bellville, 7535, RSA; 2001084@uwc.ac.za
**Faculty of Engineering, University of Technology, Sydney, NSW 2007, Australia;
agbinya@eng.uts.edu.ua

Abstract—Developers of SIP mostly concentrated on making
SIP to do what no other protocol can do. In a way they failed to
consider the weaknesses and dangers. From this realization we
did a study on the weaknesses of SIP and SIP signaling in
Wireless Networks and Services. From our study we identified
Delay and Security as the most severe problems, which we
discuss in this paper.

Index Terms—Delay, Quality, Network Security, QoS, SIP.
I. INTRODUCTION
IP is currently receiving much attention and seems to be
the most promising candidate as signaling protocol for the
current and future IP telephony services, which is becoming a
real competitor to plain old telephone service. For the
realization of such a scenario, there is the obvious need to
provide a certain level of quality and security, comparable to
that provided by the traditional telephone system. We attempt
to summarize the cost of using SIP signaling on wireless
networks and devices. From the study, we identified security
[3] and delay [1] as the most severe problems in SIP
Signaling. Different types of security and delay, are discussed
in Section II and III respectively. In section IV we discuss
VoIP.
II. QUALITY AND SECURITY
Although security and privacy should be mandatory for an IP
telephony architecture, most of the attention during the initial
design of the IETF IP telephony architecture and its signaling
protocol, SIP, has been focused on the possibility of providing
new dynamic and powerful services, and simplicity. Less
attention has been paid to security features.
If the new IP telephone architecture and SIP is to replace
PSTN, it should provide the same basic telephony service with
a comparable level of QoS (Quality of Service) and network
security.

The following security characteristics should be guaranteed:
1. high service availability
2. stable and error-free operation
3. protection of the user-to-network and user-to-user
network traffic



SIP messages may contain information a user or server wishes
to keep private. The headers can reveal information about the
communication patterns and content of individuals, or other
confidential information. The SIP message body may also
contain user information (media type, codec, address and
ports, etc) that should not be revealed.

Securing SIP header and body information can be motivated
by two different reasons:
1. Maintain private user and network information in
order to guarantee a certain level of privacy
2. Avoiding SIP sessions being set up or charged by
someone faking the identity of someone else.

The mechanisms that provide security in SIP can be classified
as end-to-end or hop-by-hop protection. End-to-end
mechanisms involve the caller and/or callee SIP user agents
and are realized by features of the SIP protocol specifically
designed for this purpose (e.g. SIP authentication and SIP
message body encryption). Hop-by-hop mechanisms secure
the communication between two successive SIP entities in the
path of signaling messages. SIP does not provide specific
features for hop-by-hop protection and relies on network-level
or transport-level security. Hop-by-hop mechanisms are
needed because intermediate elements may play an active role
in SIP processing by reading and/or writing some parts of the
SIP messages. End-to-end security cannot apply to these parts
of messages that are read/written by intermediate SIP entities.

Two main security mechanisms are used with SIP:
authentication and data encryption. Data authentication is used
to authenticate the sender of the message, and to ensure that
some critical message information was unmodified in transit.
This is to prevent an attacker from modifying and/or replaying
SIP requests and responses. Data encryption is used to ensure
confidentiality of SIP communications, letting only the
intended recipient decrypt and read the data. This is usually
done using encryption algorithms such as Data Encryption
Standard (DES) and Advanced Encryption Standard (AES).

SIP supports two forms of encryption:
1. end-to-end
2. hop-by-hop
End-to-end encryption provides confidentially for all
information. On the contrary, hop-by-hop encryption of whole
S

Page 2

173

2
SIP messages can be used in order to protect the information
that should be accessed by intermediate entities, such as From,
To, and Via headers. Encryption of such information can
prevent malicious users from determining who calls who, or
accessing route information. Hop-by-hop encryption can be
performed by security mechanisms external to SIP.

The security mechanisms must be combined properly to obtain
a trusted network scenario. SIP communications are
susceptible to several types of attacks:
1. The simplest attack in SIP is snooping, which permits
an attacker to gain information on users identities,
services, media, network topology and so on. This
information can be used to perform other types of
attacks.
2. Modification attacks occur when an attacker intercepts
the signaling path and tries to modify SIP messages
in order to change some service characteristics. This
kind of attack depends on the kind of security used.
3. Snoofing is used to impersonate the identity of a server
or user to gain some information provided directly or
indirectly by the attacked entity.
4. Finally, SIP is especially prone to denial of service
attacks that can be performed in several ways, and
can damage both servers and user agents.
The attack techniques may cause memory exhaustion,
processor overload, and so on.

Although the security mechanisms provided with SIP can
reduce the risk of attacks, there are some limitations in the
scope of the mechanisms that must be considered.
III. DELAY
The SIP-T signaling system is a mechanism that uses SIP to
facilitate the interconnection of PSTN with carrier class VoIP
network. Based on IETF, the SIP-T signaling system not only
promises scalability, flexibility, and interoperability with
PSTN but also provides call control function of MGC (Media
Gateway Controller) to set up, tear down and manage VoIP
calls in carrier class VoIP network. The performance analysis
of SIP-T signaling system plays an essential role in optimizing
network QoS (Quality of Service). Queuing size, mean of
queuing delay and the variance of queuing delay are the major
performance of MGC in carrier class VoIP network. [1]
assume a mathematical model of the M/G/1 queue with non-
preemptive priority assignment to represent SIP-T signaling
system. They also presented the formulas of queuing size,
queuing delay and delay variation for the non-preemptive
priority queue by queuing theory respectively.
Since a satellite component has been identified within
Universal Mobile Telecommunication System (UMTS), there
is a need to support SIP-based sessions over Satellite-
UMTS(S-UMTS) as well as to achieve an end-to-end
seamless IP-based terrestrial/satellite network integration. [4]
aims at evaluating the performance of SIP-based session setup
over the S-UMTS, taking into account the larger propagation
delay over the satellite as well as the impact of the UMTS
radio interface.
A. QUEUING DELAY
The queuing size was analyzed [1] using imbedded Markov
chain and Semi-Markov process and the queuing delay and
delay variation was analyzed using LST (Laplace-Stieltjes
Transform) [9]. Queuing size is defined as the number of SIP-
T messages in the system.

Mathematical
SIP-T message
(messages/s)
50 0.49
100 0.62
150 0.78
200 1.00
250 1.29
300 1.71
350 2.37
400 3.57
450 6.39
500 25.00
TABLE 1: MATHEMATICAL RESULTS [1]

Simulated
SIP-T message
(messages/s) percent (ms)
Mean (ms) Std Dev (ms) Buffer Size
0.49
0.68
0.89
1.14
1.46
1.90
2.57
3.75
6.41
18.57
0.12
0.23
0.37
0.52
0.72
0.97
1.33
1.92
3.17
8.20
Mean and 95th Std.
Dev
(ms)
0.47
0.65
0.88
1.10
1.40
1.91
2.57
3.98
6.67
20.12
Buffer
Size
Sample Size
(SIP-T
message)
4925
10074
15044
20178
25138
30115
34891
39955
45067
50251
59.2
100.7
150.4
201.8
251.4
301.1
348.9
399.5
450.7
502.5
0.49±0.01
0.61±0.01
0.79±0.01
0.98±0.02
1.27±0.02
1.70±0.02
2.36±0.03
3.62±0.04
6.49±0.06
21.08±0.18
TABLE 2: SIMULATION RESULTS [1]

Table1 and table2 shows the comparison data between
mathematical and simulation results including buffer size,
mean queuing delay and standard deviation of queuing delay
for Pu=0.004. Pu denotes the error probability of SIP-T
message.
The increase of the size of the queue, mean queuing delay and
standard deviation of queuing delay vary slowly as arrival rate
of SIP-T messages less that 450. However, the values increase
dramatically with the arrival rate of SIP-T messages greater
than 450. An intuitive and reasonable explanation for this
phenomenon is that the SIP-T message arrival rate approaches
the processing capability of the system for heavy traffic
intensity. [1]
From the results in table1 and table2, we can see that the
theoretical estimates are shown to be in excellent consistence
with simulation results.
0.12
0.25
0.39
0.54
0.76
1.03
1.40
1.97
3.49
8.85
B. CALL SETUP DELAY
The SIP protocol enables a wide set of applications and
Multimedia over IP (MoIP) is one of them. SIP-based video-
telephony is one of the most challenging Multimedia over IP
applications. A test in [2] has been carried out in a 3G
network emulator, to measure post-dialing delay, answer-
signal delay and call-release delay. These delays happen
during the lifetime of a SIP call. This results were compared

Page 3

173

3
to local, national, international and overseas Intranet LAN
calls.

Post-Dialing Delay (PDD) is also called post-selection delay
or dial-to-ring delay. This is the time elapsed between when
the caller clicks the button of the terminal to call another caller
and the time the caller hears his terminal ringing.

Answer-Signal Delay (ASD) is the time elapsed between
when the callee picks-up the phone and the time the caller
receives indication of this.

Call-Release Delay (CRD) is the time elapsed between when
the releasing party (the caller) hangs-up the phone and the
time the same party can initiate/receive a new call.

These delays are shown in the following table:
Local
SIP Call SIP Call
PDD 24ms 38ms
ASD 23ms 31ms
CRD 11ms 30ms
TABLE 3: RESULTS OF PDD, ASD AND CRD FOR 3G SIP CALLS VS.
INTRANET CALLS [2]

Tests incase of bandwidth limitations has been done [2]. The
call success rate was always 100%, thus bandwidth would not
stand in the way for a successful SIP call.

PDD for 2kbps bandwidth was less that 1 second. PDD for
5kbps bandwidth was approximately 420ms. Values of PDD
decrease with increasing bandwidth. At the maximum
bandwidth of 254kbps, the PDD was around 50ms. In realistic
3G network configurations, the bearer allocated for signaling
could be a few kbps. [2]

ASD values were constantly 45ms for channels of at least
5kpbs, but ASD increased for very narrow channels (2kpbs to
166 ms). We think this due to the fact that loss is much higher
for narrow bandwidth, whereby a 200/OK must be
retransmitted. [2]

CRD could not be measured, because the media packets were
queued up in the simulator, and they blocked the channel for a
long time. [2]

Globally, these SIP signaling values are well in line with the
Grade of Service (GoS) bounds proposed by the ETSI
TIPHON QoS classes [10]. Air interface losses and narrow
channels have a great impact on the overall SIP call setup
time. This yields large call setup times. However, it is
expected that UDP/IP header compression and SIP message
compression algorithms would greatly reduce the SIP call
setup delay over 3GPP networks.
National International
SIP Call
153ms
147ms
138ms
Overseas
SIP Calls
24ms
237ms
230ms
3G
Calls
62ms
45ms
50ms
SIP
C. MESSAGE TRANSFER DELAY
Tests done in [4] were for message transfer delay for the SIP
INVITE at block error rates of 0%, 10% and 20%. Since the
INVITE method is deemed to be the most important method in
SIP, as it is the only method used to establish a session
between participants and normally contains the description of
the session to be setup, it is interesting to ascertain its
performance.
It is found that the message transfer delay increases as the
message size increases. This is expected. It is found that the
transfer delay of the INVITE request is substantially reduced
compared to when no link-layer retransmission is employed,
whereby the delay reduction is more as the message size and
the block error rates increases. This is because without
retransmission at the link-layer, a segment that is lost means
that the whole message cannot be recovered at the receiver
side and thus, the whole message needs to be retransmitted at
the session-layer, according to the SIP reliability mechanism.

The tests also show that the delay is lower when having
unsolicited and solicited STATUS report option set, compared
to only having the solicited feedback. This is because by
incorporating unsolicited feedback on top of solicited, the
missing protocol data units (PDU) can be recovered faster
since retransmissions of missing PDUs can be performed
before polling; also the reduction in delay is more at a higher
block error rate (BLER).
D. SESSION SETUP DELAY
From tests done [4], it was shown that, session setup delay
and call blocking probability for a simple call setup sequence,
whereby the call setup consists only of sending of the INVITE
request and the 180 ringing response. Results were presented
for the different STATUS report triggers settings with Tgood
ranging between 0.5s and 10s, and Tbad equals to 0.5s, 2s and
4s. Comparing both schemes, it can be seen that when the
channel is good, i.e. for a lower Tbad value a higher Tgood
value, there is hardly any difference in performance, but as the
channel gets worse, combining both unsolicited and solicited
feedback options gives a lower delay and blocking
probability.

Due to the inherent characteristics of SIP signaling being
transactional-based and generous in size, the transport of these
packets over the radio interface is not sufficient and when
transversing over the error-prone wireless link plus a larger
satellite propagation delay, the session establishment delay
can be rather large. In our opinion SIP over Satellite is not
advisable. From the study in [4], they have shown that with
the presence of RLCAM, the session setup performance can
be substantially improved. Also it is shown that the
combination of unsolicited and solicited STATUS report
trigger gives a better performance than just solicited alone in
terms of delay and blocking probability in a more hostile
environment.
E. HANDOFF DELAY
[6] define total handoff delay (D) as the time between
detachment from old access medium and establishment of
communication with correspondence node (CN). It consists of
three (3) components:
1. time for switching lower layer medium to access
network (D1).
2. time for detecting a new router and a new link (D2)

Page 4

173

4
3. time for recovery of communication with a CN after
detecting a new link (D3).
D3 is the one that [6] concentrated on. There are two main
factors which contribute to delay D3; Duplicate Address
Detection (DAD) and router selection. DAD imposes delay
between receiving a Router Advertisement (RA) and sending
packet out of the interface with auto configured IPv6 address.
The purpose of DAD is to confirm the uniqueness of the IPv6
address on the link.

[6] measured the handoff delay of SIP terminal mobility in
their IPv6 testbed. Two different scenarios have been
considered:
a) SIP mobility without kernel modification
b) SIP mobility with kernel modification.
Measurement has been done for scenario a) and b). The
following table shows the handoff delay, D3, which is related
to signaling:
Handoff case a)
H12 38290.0 ms
H23 3932.2 ms
H31 1934.7 ms
TABLE 4 HANDOFF DELAY OF SIGNALING [6]

The following table shows another handoff delay related to
voice communication and the results is shown:
Handoff case a)
H12 38546.3 ms
H23 4187.7 ms
H31 1949.4 ms
TABLE 5: HANDOFF DELAY OF MEDIA UDP PACKET [6]

We observed from tables4 and table5, modified kernel has
reduced handoff delay. We understand that even with the
modified kernel, the delay figures are not acceptable for real-
time multimedia communications. To complete the study in
[6], they then integrated the MIPL MIPv6 in their testbed and
performed the same experiment as done for SIP mobility.
They observed that the handoff delay for signaling is about
2ms and the handoff delay for media UDP packet is less than
31ms. The results showed that MIPL MIPv6 with modified
kernel outperforms the SIP mobility with modified kernel.
However, they believe that application layer mobility, such as
SIP mobility, is a potential candidate to support real-time
applications.
b)
171.4 ms
161.6 ms
161.1 ms
b)
420.8 ms
418.6 ms
408.4 ms
IV. FAX OVER IP
For new Voice over IP (VoIP) networks to succeed, they will
have to support legacy fax equipment installed at user
premises worldwide. This fax requirement is attached to the
PSTN, which is very robust and reliable. However, the reach
and the popularity of the Internet combined with the fact that
its use is almost free, is a major driver for VoIP networks and
for the transfer of all applications that exist on the PSTN, to it.
The most major of these applications is fax transmission.
Fax transmission has special requirements, the first being that
while the loss of a packet during a human conversation is not
likely to affect a voice call a lot, it can easily affect a fax call.
This is because fax transmission requires far more signaling
and handshaking than a regular telephone call. This includes
negotiating details such as speed, paper size, and delivery
confirmation. Apart from the signaling in a fax call, the
sending and receiving of fax documents are mostly done and
interpreted by automated fax machines. Therefore, any error
in either the signaling or the actual transmission is likely to
result in a lengthy recovery.

For three network models, [8] measured the link utilization of
the inter-proxy-server link (IPSL), the link utilization of the
auxiliary link (AUXL), the average end-to-end delay of SIP
signaling packets, the average end-to-end delay of fax data
packets, the average SIP call setup time, and the average fax
call setup time. The three network models are discussed
below:

Network Models:
1. calls are generated from the T.38 gateway, and all
messages are sent to the SIP proxy server on its
network. The path between the T.38 gateway and the
SIP proxy server has an IP router on it. This is not
necessary, but is done to maintain compatibility with
the next two network models where IP routers play
an important role. The SIP proxy server of the
originator’s network communicates with that of the
recipient’s network and initially sends SIP messages
to set up a call. Once that is done, the T.38 gateway
starts fax transmission. The originating T.38 gateway
starts sending fax packets to the SIP proxy server.
That is, the SIP proxy server routes and interprets
SIP messages. But it also routes IFP packets without
actually interpreting them. Such a scenario is likely if
the SIP proxy server is implemented on a router
within the network. In that case, it interprets SIP
messages, possibly translating them, and maintaining
state for them. But all other messages are not
interpreted by it – they are just routed. This model is
shown in figure1.


FIG. 1: SIP PROXIES ROUTING SIP AND FAX PACKETS[8]

2. has the same network elements but a different setup. It
includes the components and links of the first model,
but also has a direct link between the IP routers. That
is, all the SIP signaling is carried out on the path that
traverses the two SIP proxy servers. However, once
the call is setup, all the fax data packet transmission
is done through IP routers and links only. This is
more likely to be a network where signaling travels
on separate links, and data is sent across another set
of links. This is possibly because data does not need
to go through SIP proxy servers or other network
entities. This frees up resources at such entities and
segregates signaling from data transmission, much
like PSTN networks where the Signaling System #7

Page 5

173

5
(SS7) links are distinct from trunks. Figure2 gives a
view of this model.


FIG. 2: SEPARATE PATHS FOR SIP MESSAGES AND FAX
IMAGES [8]

3. uses the same network as network model 2. However,
not all SIP messages travel between the SIP proxy
servers in this case. In general, all SIP terminals are
configured to know their network’s SIP proxy server.
Hence, to set up a SIP call, they contact their SIP
proxy server first and do not need to know the entire
route to a receiver themselves. They only need to
know how to route SIP calls they initiate to their
designated SIP proxy server. The SIP proxy server
then handles all the signaling for the rest of the call.
For fax data packets, the AUXL link is used as is
done in network model 2.



Average link utilization of IPSL
(Mbps)
Average link utilization of
AUXL (Mbps)
Average end-to-end fax data
packets delay (sec)
Average end-to-end signaling
delay (sec)
Average SIP call setup times
(sec)
Average fax call setup time
(sec)
Model 1
77.50
Model 2
9.223
Model 3
0.9546
69.03 75.96
0.2023 0.2017 0.2017
0.2014 0.2015 0.2009
3.4023 3.4042 3.4020
4.2086 4.2106 4.2060
TABLE 5: EXPERIMENT RESULTS [8]

In terms of just link utilization values on IPSL, network model
3 is the most suitable. In terms of just link utilization values
on AUXL, network model 2 is the most suitable, but it is only
marginally better. In terms of end-to-end fax data packet
delay, network models 2 and 3 are equally good. In terms of
just packet end-to-end delay, network model 3 is the best. In
terms of just average SIP call setup time and average fax call
setup, network model 3 is the best.
V. CONCLUSION
In this paper the most severe problems, security and delay, of
SIP and SIP signaling in Wireless Networks and Services are
discussed. We found that though there are security
mechanisms provided with SIP, that can reduce the risk of
attacks, there are still some limitations in the scope.
There are about five delay types we found, namely: Queuing
Delay, Call Setup Delay, Message Transfer Delay, Session
Setup Delay, and Handoff Delay, whereby Handoff Delay is
the most severe delay. Each delay case where modified to
support real-time applications, but there is still room for
improvement, especially for Handoff Delay.
The theoretical performance analysis of SIP-T signaling
shows the robustness regardless of the effect of traffic
intensity and error probability of SIP-T message for the non-
preemptive priority queue. Thus we can determine how much
the ratio of cost to performance can tolerate, how the planning
and design is needed so as to meet the requirements of carrier
class VoIP network.
This conclude that SIP and SIP signaling will soon be on even
a better security and delay level then PSTN or any other used
protocol.
ACKNOWLEDGMENT
The authors would like to thank Telkom COE (Centre of
Excellence) and the Department of Computer Science for
giving them this opportunity.
REFERENCES
[1] J.S. Wu and P.Y. Wang, “The Performance Analysis of SIP-T Signaling
System in Carrier Class VoIP Network”, 17th International Conference
on Advanced Information Networking and Applications (AINA’03),
IEEE March 2003, pp.39

[2] I.D.D. Curcio and M. Lundan, “SIP Call Setup Delay in 3G Networks”,
Seventh International Symposium on Computers and Communications
(ISCC’02), IEEE July 2002, pp.835

[3] S. Salsano, L. Veltri and D. Papalilo, “ SIP Security Issues: The SIP
Authentication Procedure and its Processing Load”, Network IEEE
Nov/Dec 2002, vol. 16, Issue 6, pp.38-44

[4] V.Y.H Kueh, R. Tafazolli and B. Evans, “Performance Evaluation of
SIP-based Session Establishment over Satellite-UMTS”, Vehicular
Technology Conference 2003 (VTC 2003-Spring). The 57th IEEE
Semiannual, IEEE 22-25 April 2003, vol. 2, pp.1381-1385

[5] J.W Jung, R. Mudumbai, D. Montgomery and H.K. Kahng,
“Performance Evaluation of Two Layered Mobility Management using
Mobile IP and Session Initiation Protocol”, Global Telecommunication
Conference 2003 (GLOBECOM’03), IEEE 2003, Vol. 3, pp.1190-1194

[6] N. Nakajima, A. Dutta, S. Das and H. Schulzinne, “Handoff Delay
Analysis and Measurement for SIP based mobility in IPv6”,
International Conference on Communications (ICC’03), Vol.26, IEEE
May 2003, pp.1085-1089

[7] N. Banejee, K. Basu and S.K. Das, “Handoff Delay Analysis in SIP-
based Mobility Management in Wireless Networks”, International
Parallel and Distributed Processing Symposium (IPDPS’03), IEEE April
2003, pp.224b

[8] U. Choudhary, E. Perl and D. Sidhu, “Using T.38 and SIP for Real-Time
Fax Transmission Over IP Networks”, 26th Annual IEEE Conference on
Local Computer Networks (LCN’01), IEEE November 2001, pp.74

[9] R.B Cooper, “Queues Served in Cyclic Order: Waiting Times”, The Bell
System Technical Journal, Vol.49, No.3, March 1970, pp.399-413

[10] ETSI Telecommunications and Intranet Protocol Harmonization Over
Networks (TIPHON), End to End Quality of Service in TIPHON
System; Part2: Definition of Quality of Service (QoS) Classes, TS101
329-2 v.1.1.1, July 2000