Modeling Fault Tolerant and Secure Mobile Agent Execution in Distributed Systems - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 141

As seen from Brocklehurst et al. (1994) and Jonsson (1997), it is reasonable to assume exponential
distribution of the effort to next breach, so we have the probability of breach at host i ,
P(breach at host i )

= P(breach at time t i + T I )

= P(breach at effort k i t i )

= 1 - exp(-vk i t i ) , v is a constant

= 1 - exp(- λ i t I ) , λ i = vk i

We may call v the coefficient of vulnerability of the agent. The higher the v, the higher is the prob-
ability of breach to the agent. Therefore, the agent security E would be the probability of no breach at
all hosts, that is,

n

∑

n − t

t i

∏

E = e

− t

= e =

t i i 1

i = 1

Suppose that we can estimate the coefficients of malice k i 's for hosts based on trust records of hosts,
and also estimate the coefficient of vulnerability v of the agent based on testing and experiments, then
we can calculate the desired time limits T i Ti's to achieve a certain level of security E . Conversely, if users
specify some task must be carried out on a particular host for a fixed period of time, we can calculate
the agent security E for the users based on the coefficients of malice and vulnerability estimates.

eVALuATIon r eSuLTS And Inf Luen Ce of The size of the agent

We evaluate transactional agents in terms of access time compared with client-server model. The
computation of mobile agents is composed of moving, class loading, manipulation of objects, creation
of clone, and commitment steps. In the client-server model, there are computation steps of program
initialization, class loading to client, manipulation of objects, and two-phase commitment.
Access time from the time when the application program starts to the time when the application
program ends is measured for agents and the client-server model. Figure 4 shows the access time for
a number of object servers. The non-fault tolerant and secure mobile agents show that mobile agent
classes are not loaded when an agent A i arrives at an object server. Here, the agent can be executed after
Aglets classes are loaded. On the other hand, the fault tolerant and secure mobile agents mean that an
agent manipulates objects in each object server where mobile agent classes are already loaded, that is,
the agent comes to the object server after other agents have visited on the object server. As shown in
Figure 4, the client-server model is faster than the transactional agent. However, the transactional agent
is faster than the client-server model if object servers are frequently manipulated, that is, fault tolerant
and secure mobile agent classes are a priori loaded.
A simulator was designed to evaluate the algorithm. The system was tested in several simulated
network conditions and numerous parameters were introduced to control the behavior of the agents.
We also investigated the dynamic functioning of the algorithm. Comparing to the previous case, the
parameter configuration has a larger effect on the behavior of the system. The most vital parameter was
the frequency of the trading process and the pre-defined critical workload values.