Вы находитесь на странице: 1из 13

AAA and PKI in Ad Hoc Networks

Pradip Lamsal
Helsinki University of Technology
Telecommunications Software and Multimedia Laboratory
pradip.lamsal@helsinki.fi

Abstract

Both AAA and PKI are frameworks that address security concerns in a network.
AAA provides a mechanism for authentication, authorization and accounting whereas
PKI provides a mechanism for managing cryptographic keys. The use of these frame-
works is limited mainly to networks with fixed infrastructure. Their use in networks
with no fixed infrastructure (or ad hoc networks) is still a research topic. In this paper
we propose some ideas for using these technologies in ad hoc networking environ-
ment. The ideas that we propose here are still at the conceptual level and require
more work to make them feasible.

KEYWORDS: ad hoc networks, PKI, AAA

1 Introduction

For the most of the twentieth century the telcommunications technology was limited to
wire line telephony system. Since the start of the ’90s mobile phones have become widespread
and this started changing the telecommunications network topology. In this mobile com-
munications only the terminal devices are mobile, the infrastructure remains fixed. An ad
hoc netwok [3] employs a different networking topology, where both the terminal devices
and the infrastructure are mobile. In order words, all the devices involved in an ad hoc
network have to work both as a terminal device and a part of the infrastructure. The ad
hoc network can be formed randomly by two or more devices and any device can enter and
leave the network at their own will.

With the advancement of telecommunications technology the issues associated with it have
also advanced. One of the fundamental issues in telecommunications domain is security:
security of both the infrastructure (including terminal devices) and information it carries.
With the potential merger of telecommunications and networking technologies the issue of
security becomes even more important. The same network can now potentially carry not-
so-important chat between friends to highly sensitive financial data and military secrets, in
some cases. The security issues that exist in wire line telecommunications are hightened
in the wireless telecommunications (only terminal devices mobile) because of the vulnera-
bility of wireless links between the terminal devices and the fixed infrastructure. In ad hoc
networks, this vulnerability of wireless links becomes more serious due to the nature of ad

1
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

hoc network topology. All devices are wirelessly linked to each other and each of them are
equally prone to all the wireless security threats, including physical theft of the devices.

There have been a lot of work done to solve a wide variety of security problems. A lot
of research have resulted in several security solutions, which address different security
problems. There are two solutions that we would like to mention here: Authentication,
Authorization and Accounting (AAA) [5] and Public Key Infrastructure (PKI) [4]. AAA
is a framework which enables collection of authentication, authorization and accounting
data to a centralized server. PKI, on the other hand is a framework, which allows proper
management of cryptographic keys. At the moment the widespread use of these two secu-
rity solutions is limited to networks with fixed infrastructure, where a centralized device
is readily available or can be allocated to act as an AAA or a PKI server. It is still unsure
whether these frameworks can be ported to ad hoc networks with no fixed infrastructure.
In this paper, we present some design proposals to use AAA and PKI in ad hoc networks.

The rest of the paper is organized as follows: Section 2 contains our perception of network
topologies, seciton 3 contains a short introduction to AAA and PKI, section 4 and section
5 have short description of these technologies in different types of networks and in section
6 we propose our solution.

2 Network Topologies

In order to understand why ad hoc networks pose difficulties in using AAA and PKI we
need to understand the topology of ad hoc networks and how ad hoc networks differ from
other networks. In this section we look at the network topologies. We divide networks into
three different categories.

1 Fixed Infrastructure Network: In a fixed infrastructure network topology, the in-


frastructure is fixed and the terminal device can be either fixed or mobile. The wire-
line telephony system, mobile telecommunications such as GSM, UMTS, and mo-
bile IP are some examples of fixed infrastructure networks.

2 Ad Hoc Network: In an ad hoc network, there is no fixed infrastructure at all. All


the devices are mobile and they can all be terminal devices. A nework is formed
when two or more devices form a connection.

3 Semi Ad Hoc Network: A semi ad hoc network consists of two subnetworks: an ad


hoc subnetwork and a fixed subnetwork. The ad hoc subnetwork is connected to the
edge of the fixed subnetwork. For instance, Wi-Fi [2] is an example of semi ad hoc
network because it consists of an ad hoc subnetwork and a fixed subnetwork. The ad
hoc subnetwork accessess the fixed subnetwork via an access point.

3 AAA and PKI

In this section we describe the fundamental principles of AAA and PKI and explain how
these technologies are used in conventional wireline communications.

2
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

3.1 AAA

AAA is a framework that enables authentication and authorization of a user accessing a


network and also allows for collection of accounting information. When a user wants to
access a network, it sends a request for access to the AAA client. AAA protocol does not
specify the protocol between the user and the AAA client. Once the AAA client receives
such a request it sends an AAA request to the AAA server on behalf of the user. The
AAA server authenticates the user by evaluating the request and responds the AAA client
with proper authorization. The AAA server is also capable of querying the AAA client for
accounting information.

AAA framework consists of three fundamental compoments: AAA server, Application


Specific Modules (ASMs) and Repository. This is shown in figure 1.

Client AAA Server AAA server

ASM Repository

Figure 1: AAA Architecture

The generic AAA server has rules to evaluate the request and to make decisions about
authentication and authorization. However, this server is generic and the requests are so
application specific that the server delegates these requests to the ASMs. All the events are
logged in the policy and event repository. This repository can be used to evaluate further
requests and to access accounting information for a specfic user.

An exmaple of AAA framework is diameter [9] base protocol. This base protocol provides
the basic framework and is supplemented by AAA Transport Profile [1] and AAA Appli-
cations [7, 8]. The transport profile addresses the transport issues whereas the applications
address the application specific issues. The base protocol, as it is, can only be used for ac-
counting whereas for authentication and authorization the base protocol must be extended
for a particular application. A diameter client must support the base protocol and any
one application specific protocol whereas any diameter server must support both the base
protocol and all the application specific protocols. Without supporting all the application
specific protocols the server is unable to provide services for those applications.

3.2 PKI

PKI is a framework that is aimed at supporting the management of public keys. A public
key and its corresponding private key are used to encrypt and decrypt messages. The public

3
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

key is generally available in public whereas the private key is owned (or at least assumed
to be owned) only by the real owner. PKI uses certificates to distribute public keys. A
certificate maps a person’s identity to his public key.

Trust is a very important aspect of PKI. When someone uses a public key they have to
trust the Certificate Agency (CA) issuing that public key or the certificate containing that
public key. There are two types of trust models used in PKI frameworks: hierarchical
trust model and web-of-trust trust model. ITU’s X.509 [4] is based on hierarchical trust
model and Pretty Good Privacy (PGP) [11] is based on web-of-trust trust model. Detailed
information on the PKI trust models can be found in [6].

4 AAA and PKI in Fixed Infrastructure Networks

Both in AAA and PKI, it is assumed that a centralized server handles the core functionality.
In networks with fixed infrastructure it is feasible to dedicate a seperate server to handle
this functionality. A lot of networks already have PKI framework in place. Adding AAA
framework is not that technically challenging.

5 AAA and PKI in Semi Ad Hoc Networks

From the perspective of AAA and PKI semi ad hoc networks are more or less very similar
to the networks with fixed infrastructure. This is because both AAA and PKI functionality
can be handled in the fixed subnetwork of the network. Both servers can reside at the
access point and the nodes in the as hoc subnetwork can use their services via the access
point. After being authenticated to the whole network the mobile nodes can be a part of
the ad hoc subnetwork and enjoy the services.

6 AAA and PKI in Ad Hoc Networks

In this section we look at how AAA and PKI can be used in ad hoc domain.

6.1 Issues with Centralized Server

As mentioned earlier, there is no fixed infrastructure at all in an ad hoc network. Since any
mobile terminal can come and go at any time, the idea of a teminal which is available all
the time does not fit in the paradigm. Also, the terminals in the network are very mobile.
In fact, they can be so mobile that they can easily go beyond the boundary of the network.
For this reason the concept of a centralized server poses a major issue in ad hoc networks.

4
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

6.2 Proposals

There have not been a lot of implementations of AAA and PKI in ad hoc networks. There
are still a lot of research conducted on different aspects of ad hoc networks. Although PKI
has been used in other networking topologies its use in ad hoc domain still raises several
issues. In this section we consider a few options for porting AAA and PKI in ad hoc
networks and some concerns about those technologies. These options are at the conceptual
level and they do not describe any implementation level details. For each of the options we
list its advantages and disadvantages.

Elected Server Model

This model borrows the concept of centralized server where the participants elect one of
them to act as the main server for the session. The network also consists of a standby
server and this standby server takes over if the main server suddenly disappears from the
network. The participants can employ any mechanism to elect the main server. One simple
mechanism is to elect the oldest participant to be the main server and the second oldest to
be the stanby server. This is shown in figure 2.

One request for


the whole network
Main
Server

New
Participant
Standby
Server

Ad Hoc Network

Figure 2: Elected Server Model: Conceptual View

In this model, both AAA and PKI functionality must exist in all the participants. The main
server is responsible for authentication, authorization and also needs to collect and store
the accounting information. It has to poll other participants for accounting information.

The trust model employed in this proposal is two-level hierarchical model. The main server
behaves as a gateway for trust meaning that if a participant is trusted by the main server then
the network (all other participants included) automatically trusts that participant. Inside the
network all participants have complete trust among each other.

The lifecycle of a participant inside the network can be divided into three categories. First
phase is when a participant enters the network. In this phase this new participant is au-
thenticated by the main server. The main server authorizes this new participant to use the
resources in the network. This authorization is based on the authorization rules of the main
server, which in fact is the AAA server. In the second phase of the lifecycle, this partici-
pant become a part of the network. In this phase it has equal trust relationship with other
participants in the network, meaning no further authentication is required. Of course, what

5
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

this new participant can do inside the network is determined by what it is authorized to do
by the main server. The last phase is involves leaving the network. A participant can either
leave gracefully when it informs the main server that it is leaving the network or quit acci-
dently without informing the network (or without informing any other participant). When
a participant leaves the network, its previous existance disappears. If it wants to rejoin the
network, it has to start from the first phase.

The lifecycle for the main and standby servers is a bit different because of the fact that
they do a lot more than the other participants. When the main server leaves the network
gracefully, it has to handover its reponsibility to the standby server. The standby server, in
turn, organises an election to select a new standby server. However, if the main server quits
accidently then the standby server needs to take over as the main server and the network
can then elect a standby server. For this, the standby server has to constantly monitor the
main server and be prepared to be the main server at any time.

Protocol

The protocol, which is used by the client to access the network, consists of request and
response messages. This is shown in figure 3.

Client Server Server2


Request

Response

Accept
Session key
Session key accepted
Trust
cache

Broadcast client info

Figure 3: Elected Server Model: Protocol Overview

The client requests the main server for the access. The client specifies what resources
it would like to access and what kind of access it would like. The main server does the
evaluation and sends response to the client. This response indicates the client what kind
of access the server can give to that client on behalf of the network. The client can either
accept or reject the response. If the client decides to accept what ther main server offers
then the client and the main server can exchange a session key. This session key then
becomes a token of trust between the new participant and the network. This trust is then
stored in the trust cache for further use, if necessary. The trust cache is local to all the
participants. Once the client becomes is accepted in the network, the main server sends the
client’s information to all other participants so that those participants have the up to date
information of the network.

Next we discuss some advantages and disadvantages of this model.

Advantages

i Everything is centralized and accounting information can be easily obtained by

6
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

querying the main server.

Disadvantages

i Each mobile terminal node must act as an AAA and PKI server. This is quite de-
manding considering the fact that the mobile devices have limited computational
power, limited memory, limited battery life etc. This makes all the terminals very
bulky in terms of the extra functionality they are expected to support for the network.
Most of the time and for most of the terminals this extra functionality is hardly used
for a given session.

ii From the trust point of view, this is a very dangerous model. Once a mobile terminal
enters the network there is nothing stopping it to become the server at a later stage
given that it enters the network at the right time. Once it becomes the server, it has
equal authority and the rest of the participants are expected to trust it completely.
Now, if this terminal happens to be an advarsary and somehow enters the network,
the advarsary terminal can have the control of the entire network.

iii The handover mechanism is very weak. All the participants have to constantly mon-
itor if the server is still in the network or not. Not only that, they all have to be aware
when standby server takes over and when that happens, they have to elect another
participant as the new standby server. This makes all the participants a potential
server.

iv The exit from the network is very difficult to track. Since the main server cannot
keep track of all the participants they should all have some kind of session specific
information which expires once a participant loses connection with its last contact.

All Server Model

This model does not have the concept of "the server" but all the participants become a
server of their own. Each of them is responsible of taking care of itself and they are all
independent of each other. This is illustrated in figure 4.

One request per


mobile device

New
Participant

Ad Hoc Network

Figure 4: All Server Model: Conceptual View

7
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

In this model both AAA and PKI functionality must exist in all the participants. All the
participants act as both the server and the client. When acting as the server, each partici-
pant must authenticate its clients, which are requesting access. The server gives adequate
authorization indicating what each client is allowed to do. The server also records the
accounting information for each of the clients. When acting as a client, each participant
should request for authentication and authorization from the server.

This model employes the web-of-trust trust model. The trust is per device and trust between
any two participants does not extend beyond those participants. The access of a participant
is limited only to those servers with which it has established trust.

The lifecycle of a participant can be divided into three phases. The first phase is when
the participant enters the network. Entering the network here refers to having connection
to at least one of the participants of the network. In this phase, the new participant is
authenticated by the server it is trying to access. The server authenticates this participant.
After a successful authentication the participant is granted authorization and it can access
the server. The second phase is after the participant enters the network. In this phase
the participant can access the servers with which it established connection earlier. The last
phase of the lifecycle involves leaving the network. It is not important whether a participant
leaves the network gracefully or quits accidently. As soon as the the connection is cut off,
rejoining the network requires starting from the first phase.

Protocol

The protocol, which is used by the client to access the network, consists of request and
response messages. This is illustrated in figure 5.

Client Server

Request

Response

Accept
Session key
Session key accepted

Trust
cache

Figure 5: All Server Model: Protocol Overview

The client requests each of the servers for the access. The basic mechanism is quite similar
to the electer server model. A session key is established for each server and that session
key is valid only for that server. This trust is stored on each server for each client and can
be used in the future, if required.

Advantages

i This model has a better security model than the previous one. This is because if a
device is compromised, this does not compromise the whole network as access to a
device does not give access to other devices automatically.

8
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

ii The last phase of the lifecycle can be desgined to allow for better caching. It is
possible to make the server ’remember’ its clients for a certain amount of time if the
clients leave gracefully. This can help the server establish stronger trust relationship
with the same client next time.

Disadvantages

i This model is more like peer-to-peer computing model rather than a network.

ii If a network contains a lot of participants, the overhead of authentication and autho-


rization is quite big.

iii This model contains all the issues related to the limited power of mobile terminal
devices such as computational power, memory, battery life etc.

iv Accounting information of a device is distributed across the network. If an external


entity wants to collect this accounting information, it has to poll all the devices. This
can be quite a waste of the bandwidth.

Group Server Model

This model has the concept of server but the functionality of the server is shared among
a set of devices in the network. This model is based on the theory behind threshold cryp-
tography [10]. Out of M (=m+n) mobile device any m devices, as a group, can act as the
server for the network. This is illustrated in figure 6.

One request for


the whole network

New
Participant

Group of devices
making the server

Ad Hoc Network

Figure 6: Group Server Model: Conceptual View

Both the AAA and PKI functionalities are distributed among the mobile devices. When the
network receives a request from a new participant to join the network, that request is dis-
tributed over the m devices, which are acting as a server. Different devices are responsible
for different aspects of the functionality and the the final decision is based on the combined
result of the server. A client can send its request to join the network to any mobile device
in the network. The mobile device which receives such a request from the client becomes
the interface device for that client. This interface device delegates the requests to those
mobile devices, which constitute the server in the network. The interface device collects
the responses from other devices, which received the requests and responds to the client

9
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

with the result. The way how and where the accounting information is stored is still an
open issue.

The trust model in this group server model is more like hierarchical. However, this hier-
archical model is a bit different from the hierarchical model of the Elected Server Model
or the one described in X.509. From a participant’s point of view, the trust is established
between the participant and the mobile device it comes in contact with first. So, the par-
ticipant perceives that mobile device as the trusted third party. However, for the devices,
collectively acting as the server, the trust is between the collection of the devices and the
new participant. So, in reality, there is no single trusted third party. And, the trust estab-
lished this way is the trust between the new participant and the network irrespective of how
many devices worked together to make the server.

The lifecycle of a participant in this model is exactly the same as the one in the Elected
Server Model. The only difference here is that instead of one device acting as the server
there is a group of devices that act as the server. Because of this distinction, this model
does not have the concept of main server and standby server.

This model requires each terminal device to have some functionality of AAA and PKI
server. Here, the AAA and PKI functionality is distributed over several terminal devices
and the result is a collective work of those devices. For instance, the authentication of a
new participant can be distrubuted over three different devices and their collective result
can determine the result of authentication. This can be made more robust by delegating the
task to any m+n devices and taking the result from any m devices. This can accomodate
any possible disagreement among devices or even the possibility of n devices being taken
out of the network. Storing accounting information, however, needs careful thought.

Once a terminal device is authenticated, it has access to the whole network. Afterwards, it
can become one of the devices providing a part of AAA and PKI services to other terminal
devices. When the device leaves the network, it does not have a big impact to the AAA
and PKI services of the network.

Protocol

The protocol, which is used by the client to access the network, consists of request and
response messages. This is illustrated in figure 7

Client S1 S2 S3 S4
Request
Work1
Work2
Work3
Reply1
Reply2
Reply3
Response
Accept
Session key
Accept session key Store trust

Figure 7: Group Server Model: Protocol Overview

10
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

The client sends a request to any server in the network. The server which received the
request becomes the point of contact of the client to the network. Once this server receives
the request it distributes the task of evaluating the request to those m servers and receives
the responses. It then examines the responses and evaluates the final response and sends
it to the client. Like in the previous protocols the client can either accept or reject the
response. If accepted, the session key is established and trust is cached.

Advantages

i If a terminal device quits the network accidently, the impact of this incident to the
overall AAA and PKI functionality is minimal since this functionality can be redel-
egated to another device.

ii Distributing the functionality over several terminal devices also reduces the overall
impact of up to n devices being compromised.

iii The trust model is reasonably straightforward and is not on a per device basis.

Disadvantages

i The distribution of services becomes a significant issue if there are less participants
in the network.

ii Storing accounting information is an issue. This issue can be resolved by making


each terminal device record its resource used by other devices.

6.3 Open Issues and Future Work

All the proposed models have some common issues. These issues have less to do with the
models but more to do with the characteristics of the mobile devices. Due to the limited
capability of mobile devices in terms of processing power, memory and power require-
ments, these devices are not capable of having all the heavy functionalities on them. These
devices have to work as routers as well in addition to being terminal devices. Expecting
these mobile devices such as a mobile phone or a PDA (Personal Digital Assistant) to have
the additional functionalities can be beyond what they are designed for. If these mobile
devices start to become more powerful like laptops, such expectation seems feasible.

Having said that we believe that further studies are needed in the following areas:

1 Work can be done to slim down the diameter protocol so that it is more suitable for
less powered devices.

2 Since mobile devices have to work as routers in ad hoc networks, work can be done
to combine routing and authentication.

3 In all the models that are propsed in this paper, authentication and authorization
do not require that much of storage unlike accounting information. One important
reason for storing accounting is for payment. One area where investigation can occur
is some form of ’pre paid’ system where a participant has to pay the network during

11
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

authentication and then use that credit to use the services or resources. Depending on
the model used, pre-payment can be done either to the network or individual devices.

4 The lifecycle of a mobile device should be investigated in more details for all the
models.

5 The implementation details are missing. To consider the details some commonly
available devices can be investigated in terms of memory available, computational
power of the processor and the possibility of having trust cache.

7 Concluding Remarks

Although AAA and PKI are used in fixed infrastructure networks their use in mobile ad hoc
networks is still more into research domain. The three proposals that we have presented in
this paper are not meant provide the complete solution but they are a good starting point
for further research. These models are still more at conceptual level. All models need more
and detailed work and attention should be paid to the implementation details.

References

[1] B ERNARD A BOBA, J ONATHAN W OOD; AAA transport profile, RFC 3539; IETF
Internet Working Group 2003.

[2] W I -F I A LLIANCE; http://www.wi-fi.org .

[3] S COTT C ORSON, J OSEPH M ACKER; Mobile ad hoc networking (manet): Routing
protocol performance issues and evaluation considerations, RFC 2501; IETF Internet
Working Group 1999.

[4] ITU-T; Information technology - open systems interconnection - the directory: Pub-
lic key and attribute certificate frameworks; ITU-T Recommendation X.509|ISO/IEC
9594-8 2000.

[5] L EON G OMMANS J OHN R. VOLLBRECHT C EES T.A.M. DE L AAT, G EORGE


M. G ROSS, DAVID W. S PENCE; Generic AAA architecture, RFC 2903; IETF In-
ternet Working Group 2000.

[6] P RADIP L AMSAL; Understanding public key cryptography and PKI;


http://www.cs.helsinki.fi/u/lamsal/asgn/pki/AsymmetricCryptographyAndPKI.pdf
2002.

[7] DAVID S PENCE PAT R. C ALHOUN , G LEN Z ORN, DAVID M ITTON; Diameter net-
work access server application, IETF draft; IETF Internet Working Group 2003.

[8] JARI A RKKO E RIC G UTTMAN PAT R. C ALHOUN , J OHN L OUGHNEY, G LEN Z ORN;
Diameter base protocol, RFC 2903; IETF Internet Working Group 2003.

[9] T ONY J OHANSSON PAT R. C ALHOUN, C HARLES P ERKINS; Diameter mobile IP


application, IETF draft; IETF Internet Working Group 2003.

12
HUT TML 2003 Tik-110.557 Research Seminar on Telecommunications Software

[10] L IDONG Z HOU, Z YGMUNT J. H AAS; Security ad hoc networks; IEEE Networks
13(6):24–30; 1999.

[11] P HIL Z IMMERMAN; The Official PGP User’s Guide; MIT Press; 1995.

13

Вам также может понравиться