In the case where no AC is available from CE-2 to PE-2, a
new VC through PE-2 is established. In this case, PE-2 will
take part into OVPN-1 to establish the new VC, which is
composed of an AC from CE-2 to PE-2, an OC from PE-2 to
PE-3, and an AC from PE-3 to CE-3. If no more packet radio
channel passes through PE-1, the mentioned reconfiguration is
equivalent to an all replacement by a new serving OVPN
(OVPN-2).
The reconfiguration of a new OVPN always produces a
considerable delay. Therefore, the edge GGSNs must inform
to the OVPN providing system as soon as possible when a
packet radio channel is about to be changed. The cooperation
between the prediction methods and “early alert” will reduce
the damage caused by the reconfiguration.
III. AGENT-BASED OVPN MODEL TOWARDS GPRS ROAMING
The efficiency of dynamically providing an OVPN along
with GPRS roaming depends on the knowledge exchanging
capacity between PE devices and the individual decision
making capacity of each node. These capacities have been
proved in agent-based systems [11]. The integration of the
multi-agent technology into the OVPN model can benefit from
their capacities an features.
With the reference model of OVPN and its functions
mentioned above, three types of agents are defined for each
OVPN component: a) the Controlling agent that manages
OVPN agents, b) the Grooming agent that optimizes the usage
of network resources, and c) the OVPN agent that establishes,
maintains and releases VCs belonging to an OVPN. Figure 4
illustrates the interaction between these agents.