Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Mobile Distributed Authentication Protocol: II. Authentication in GSM Networks

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Home

Mobile Distributed Authentication Protocol


A. F. Al Shahri, D. G. Smith and J. M. Irvine
Dept. of Electronic & Electrical Engineering
University of Strathclyde, Glasgow, G1 1XW, UK
aied@comms.eee.strath.ac.uk, d.g.smith@eee.strath.ac.uk, j.m.irvine@strath.ac.uk
Abstract - Networks access control is a crucial topic and
authentication is a pre-requisite of that process. Most existing
authentication protocols (for example that used in the GSM
mobile network) are centralised. Depending on a single entity is
undesirable as it has security, trust and availability issues. This
paper proposes a new protocol, GSM-Secure Network Access
Protocol (G-SNAP). In G-SNAP, the authentication procedure
and the network access control is handled by a quorum of
authentication centres. The advantages of the novel protocol
include increased security, availability and a distributed trust.

protocol is based on secret sharing schemes, which have a


quorum access structure. Section II introduces authentication
in GSM networks. Section III describes the proposed
protocol G-SNAP. Section IV provides analysis and
simulation results. A comparison between G-SNAP and
centralised approaches is discussed in Section V.

I. Introduction

In GSM, authentication is achieved by checking the validity


of a subscribers SIM card. An authentication algorithm
(termed A3) is stored on the SIM card and also in the
authentication centre (AuC) on the network. The process is
challenge-response based. The A3 algorithm uses two input
parameters: the secret key, Ki, which is stored in the SIM
card and in the network, and a random number (RAND),
which is transmitted to the mobile station as a challenge. The
A3 algorithm uses Ki and RAND to calculate a response
(SRES). The mobile station will send back the SRES to the
network as a response to the challenge. The network uses the
same RAND, Ki, and A3 to produce an SRES, which is
checked against the response from the mobile station [5].

Authentication represents the front door of any secure


system. Strong authentication protocols are needed to restrict
network access to only authorised users. Most existing
authentication protocols are centralised, using a single
authentication centre. These centralised approaches suffer
certain drawbacks such as an attack on that single entity
compromises the whole system and if that entity becomes
unavailable then authorised users are unable to access the
system. In addition trust is focussed on that single entity
which becomes unconditionally trusted, resulting in increased
risk and it may not be suitable for the more complex business
models in 3G systems. Secret sharing schemes and quorum
systems are tools to increase security and availability and to
distribute trust between entities [1]. The basic idea of secret
sharing is to divide a secret into pieces called shares.
Thereafter, the pieces are distributed amongst users such that
the pooled shares of specific subsets of users allow
reconstruction of the original secret [2]. Non-qualified
subsets have absolutely no information on the secret. Secret
sharing schemes are useful in any important action that
requires the concurrence of several designated entities to be
initiated [3].
A quorum system for a system of nodes is a collection of
subsets (quorums) of nodes, each pair of which has a nonempty intersection. Each quorum can act on behalf of the
system. For example, let U be a set of nodes, U = {a, b, c, d}.
Then Q = {{a, b}, {b, c}} is a quorum set under U. The
intersection property enforces consistency between nodes,
which is important in replicated objects and for node coordination. Quorum systems [4] have been used for a number
of applications in the area of distributed systems including
mutual exclusion protocols, replicated data protocols,
database access control protocols and multi-party
computations. In this paper a new protocol, GSM Secure
Network Access Protocol (G-SNAP) is proposed. This

II. Authentication in GSM Networks

A. Drawbacks

A challenge-response procedure such as GSM uses is a


strong authentication process to access network
resources. The major drawback of the GSM
authentication process comes from its centralised nature.
The authentication centre in the network controls the
authentication, and as such, if the authentication centre is
attacked or compromised, then unauthorised users may
obtain access to network resources.

A centralised system also has availability issues.


Authorised users will be unable to obtain access if that
entity breaks down, denying service to users.

A final concern is one of trust. In GSM networks, the


authentication centre is unconditionally trusted. Such a
centralised trust model is not recommended, as focussing
the trust on a single entity will increase the risk, which
might affect the system security.

The communication between all users and the central


authentication node involves a lot of signalling traffic
and high load on the authentication node.

III. G-SNAP

AuC
k

G-SNAP is an extension of recently developed protocol


called a secure network access protocol (SNAP) [6]. The
proposed G-SNAP algorithm seeks to address GSM
centralised authentication deficiencies by using a distributed
approach. G-SNAP makes use of both secret sharing and
quorum structures, which have previously been used in
security applications to increase security and availability [7,
8].
In order, to perform authentication or access control in a
GSM network using G-SNAP, the mobile station contacts a
quorum of authentication centres with a quorum size k to
obtain permission. Fig. 1 depicts the mobile distributed
model. Each member of the quorum participates in the
authentication process by identifying the user and providing
the corresponding share required to complete the
authentication process. At the same time none of the quorum
members can handle the whole authentication or grant
network access on its own. In other words, the authentication
process is distributed between all the quorum members.
However, fewer than k nodes (authentication centres) cannot
grant the network access. If one of the authentication centres
becomes uncontactable, the mobile station, after a certain
time can contact another quorum to obtain the secret shares
and hence it can access the network resources. Hence to
increase availability, G-SNAP assumes that the mobile
station can contact l quorums. The selection of k and l is a
trade-off between increased security and minimised
overhead. In G-SNAP, the first (local) authentication server
(local AuC) which receives the mobile station signal will start
the authentication process. The local AuC participates in the
negotiation but does not control or take the network access
decision on its own. Fig. 2 depicts the required signalling for
the user to obtain the required shares from a quorum of k
AuCs. The G-SNAP works as follows:

S2
Sk

AuC
2

Core
Network
(CN)
RAN

S1
AC

ms
Local
AuC

AuC
1

Fig. 1 The distributed authentication infrastructure

User

Local AuC

Authentication centres

Access Request
Shares Request ?
Share Request 1
AuC 1
AuC 2
Share Request k
AuC k

Share reply 1
Share reply k

1.
2.

3.

4.
5.

6.

Top

Once the local AuC receives a signal from the ms to


register or to use the network.
The local AuC will ask the ms to obtain permission or to
be authenticated by a quorum of AuCs by providing k
shares (S1,..Sk) from any quorum.
The ms will start independently communicating with the
AuCs, which are already known by the SIM card of that
ms or given by the local AuC, and ask for the
corresponding shares. If ms does not receive a response
from one AuC in a certain time it will try another
quorum, up to l quorums.
Each AuC will check if the ms is authorised and then
will send the corresponding share if appropriate
After receiving the required k shares by the SIM card of
the ms, the ms will send the concatenated secret shares as
one message to the local AuC.
The local AuC will reconstruct the AC using a
reconstruction function. If it is correct, it will give the ms
permission to access the network or alternatively output
a rejection message.

Shares message
AC Reconstruction
Access-reply
Network access
Fig. 2: G-SNAP signalling

IV. Simulation analysis


It is important to model the performance of any new protocol
before its implementation. Most security protocols have been
evaluated from a security and information theory perspective
only. Networks are the main platform for most of the
applications including security applications. These protocols
use the networks to convey the messages between the
protocol components. Therefore, it is important to evaluate

the performance of the protocol, using metrics such as delay,


efficiency and resilience and to analyse their impact on users.
G-SNAP is developed to increase security, availability and to
distribute trust. In this paper G-SNAP has been studied using
simulations based on network simulator version 2 (NS-2) [9].
The delay experienced by users using G-SNAP or centralised
approaches is obtained. In addition a comparison between
SNAP and centralised authentication protocols is investigated

6
3
2

MSC

TABLE I
Quorum number
Q1
Q2
Q3
Q4

Mobile

Mobile

Core Network
Mobile

Mobile

Mobile

Fig. 3 G-SNAP architecture.

Top

Quorum members
1, 2, 5
2, 3, 4
0, 4, 5
0, 1, 3

User allocation
1
2,3,4
5,6
0

A. Assumptions

Mobile

Quorum system

Mobile

Q2

Fig. 4 Network topology

Mobile

Q1

Mobile

0
Q3

The simulation model involves two main steps:


The network is loaded by generating background traffic
to load all the links.
After a certain time, whenever the network has been
loaded, G-SNAP traffic will start generating the
authentication packets.
In GSM networks the AuCs are connected to the mobile
switching centres (MSCs). All the MSCs are in the core
network and are linked by a fixed network. The G-SNAP
architecture is illustrated in Fig. 3. In order to study the
performance of G-SNAP, an arbitrary core network of 7
MSC nodes is built as shown in Fig. 4 using NS-2. Fig. 4
depicts the topology of the authentication nodes using a
quorum system. For example, Q3 contains three member
nodes 0, 4 and 5. As shown in Table 1, there are four
quorums and each authentication node is a member of two
quorums. The user allocation to local quorums is shown in
the same Table. In this quorum system the quorum size k is 3.
The signalling link capacity between nodes is 1 MB/s and the
link delay is 10 ms. Each node contacts its neighbour using a
full duplex link. In this simulation each node represents an
authentication centre in the MSC.

Q4

To generate the background traffic, UDP agents are used.


Both source and destination nodes use UDP agents.
The background traffic is exponentially distributed
traffic with packet size 500 bytes.
For G-SNAP traffic ping agents are used in both the
source and destination nodes. Using the ping agent we
can calculate the round trip time delay between source
and destination.
The authentication packet size is assumed to be 250
bytes [6].
The authentication processing time is not considered in
this study.
The quorum size k is 3

B. Simulation results
To improve the statistical significance of the results, the
simulation was run for a large number of cycles. It is
assumed that each node represents an authentication centre
The user will send a packet that contains the required
information for the user to identify themselves to the other
nodes. In this case we assume that the authentication packet
is large enough to handle all the needed information. The
packet size is 256 bytes. Using G-SNAP, the user will
simultaneously send k packets to the authentication nodes.
The selection of the routing path is based on the built-in
routing protocol provided by ns-2. Since ping agents are
used, packets will go back directly after reaching the
destination towards the source. In G-SNAP the user needs to

average delay ms

90
80

G-SNAP/Q2
CAS/2

70
60
50
40
30
20
10
0
0

Users

Fig. 5 G-SNAP on Q2 and CAS on node 2

Top

60
50
40
30
20
10
0

Users 1 and 2 will experience an increased delay using


G-SNAP.
All other users will have exactly the same delay.

In the second scenario, G-SNAP users will contact local


quorums as shown in Table I. The centralised node is node 2
as before. The results are shown in Fig. 6. It appears that GSNAP performs better in this case. Users 0, 5 and 6
experienced smaller delay using G-SNAP compared to the
centralised approach. The location of the centralised node
with respect to users is a major issue.
In the third scenario, G-SNAP users will contact local
quorums and the centralised users will contact node 6. It can
be observed from Fig. 7 that G-SNAP is more efficient than
contacting a remote centralised node.
If we study these results we come up with the following
observations:
1. The delay increases if the number of hops between nodes
is increased.
2. In most of the cases some users will experience exactly
the same delay, either using a centralised authentication
centre or using G-SNAP (although traffic load will
increase).
3. The location of the authentication node with respect to
the user has a direct impact on results in both centralised
and distributed approaches.
4. Designing an efficient quorum system and distributing
the users on local quorums is crucial and can lead to
good results.

G-SNAP/local quorums
CAS/2

80
70

Users

Fig. 6 G-SNAP on local quorums CAS on node 2


G-SNAP/local quorums
CAS/6

90
80
average dealy ms

90
average delay ms

wait for the last packet to arrive such that the waiting time
does not exceed the predefined time limit. In order to
compare G-SNAP with centralised approaches, three
scenarios have been considered.
In the first one, centralised users contact a centralised
authentication server (CAS) in node 2. G-SNAP users will
contact one of the quorums. Q2 is selected. This means that
each user should contact the quorums members which are
node 2, 3 and node 4. Fig. 5 shows the average delay
experienced by users in both cases. It can be observed from
that figure the following:

70
60
50
40
30
20
10
0
0

Users

Fig. 7 G-SNAP on local quorums CAS on node 2

V. Discussion
In this discussion a comparison between G-SNAP and
centralised approaches is investigated. In order to determine
the signalling overhead caused by G-SNAP, the number of
messages sent to the authentication nodes is required. In GSNAP the user must contact k nodes to collect the required
shares. Hence, signalling overhead is increased by k
compared to centralised approaches. Contacting the local
AuC is not considered in the signalling overhead since it is
not significant. In centralised approaches all users contact a
single node increasing the traffic destined for that node. At
the same time the load on the single entity will be high. In
contrast G-SNAP distributes the load to the quorums. If there
are l quorums then the load to each quorum is equal the total
load divided by l. G-SNAP can achieve load balancing as
well. G-SNAP overcomes the drawbacks of centralised
approaches as mentioned in section II. G-SNAP increases

security since the user needs to be authenticated by a quorum


of nodes. In addition, if one node is compromised, the
attacker will only receive one share, so the system as a whole
remains secure. In the centralised case, a successful attack on
the authentication centre will compromise the entire system.
Furthermore, G-SNAP increases the availability too, if one
node or quorum is unavailable the user can contact another
quorum. Additionally, G-SNAP distributes the trust among
the quorum members and none of the quorum nodes can
grant or control access on its own. In high security
applications the advantages of G-SNAP would appear to
outweigh the additional signalling overhead.

VI. Conclusion
There is a need to have multiparty authentication protocols
such that more than one server controls the authentication
process. This results in increased security and availability,
and also distributed trust, which overcomes the centralised
authentication approaches deficiencies. The importance of
this will increase with 3G networks, which are more
distributed in nature and have more complex business
arrangements. G-SNAP is a new protocol, which achieves
these objectives. The performance evaluation of G-SNAP
compared to centralised approaches is discussed in this paper.
Although G-SNAP does introduce more signalling, the delay
impact on users is marginal. G-SNAP can be more efficient
than the centralised approach if it is used to provide more
localised distributed authentication.

References
[1] A. F. Al Shahri, D. G. Smith and J. M. Irvine, Implementation
of quorum systems to increase network security, PGNET 2002,
June 2002, Liverpol University, UK.
[2] A. F. Al Shahri, D. G. Smith and J. M. Irvine, Implementation
of Secret Sharing to Increase Network Security and Reliability,
ESPRC
Postgraduate Research in Electronics and Photonics
(PREP), April 2002, Nottingham University, UK.
[3] G. J. Simmons. An introduction to shared secret and/or shared
control schemes and their application, In Contemporary
Cryptology, The Science of Information Integrity, pages 441-497.
IEEE Press, 1992.
[4] Malkhi, Quorum systems, chapter in the encyclopaedia of
distributed computing march 1999
[5] J. Dunlop, D. Girma and J. Irvine, Digital Mobile
Ccommunications and the TETRA System, John Wiley & Sons.
1999.
[6] A. F. Al Shahri, D. G. Smith and J. M. Irvine, A Secure
network access protocol (SNAP), unpublished.
[7] L. Gong. Increasing availability and security of an
authentication service, IEEE J. Selected Areas Comm., 11(5):657662, 1993.
[8] M. Naor and A. Wool. Access control and Signatures via
quorum secret sharing, In Proc. 3rd ACM Conf. Comp. And comm.
Security, pages 157-168, New Delhi, India, Mar. 1996.
[9] Network Simulator version 2 (NS-2) UC Berkeley, USA, http://
www.isi.edu/nsnam/na/ns-documentation.

Top

You might also like