Autonomic Frameworks Deployment Using Configuration and Service Delivery Models for the Internet of Things - Interoperability and Open-Source Solutions for the Internet of Things - page 92

Fig. 2. Autonomic agent architecture diagram

agent, it monitors the IoT system. Monitoring involves multi-side monitoring: (a) IoT
monitoring: the agent will monitor IoT requirements and keep record of the perfor-
mance of the same under different scenarios monitored on the other phase. (b) Infra-
structure monitoring: the agent keeps observations on the infrastructure and help to
keep the performance parameters up to date. In this stage of the autonomic cycle the
agent will monitor the resource parameters, and relevant performance parameters. For
example, with a de ned memory capacity, the system (IoT framework) can respond to

any local request within a period of time. This performance record will be stored in the
agent memory.
The learning phase of the autonomic agent might be approached in two different
ways. The agent can be deployed empty, meaning the database keeping track of the
performance parameter against different con guration parameter can be empty at the

beginning of the deployment and it periodically learn from the system as time goes.
In this case, at the beginning of the deployment, the agent only has an initial infor-
mation for the IoT, for example what are the minimal resource requirements for the
deployment of the platform, and it is assumed that the owner of the IoT system knows
best on the issue that

“ what con guration ensures the best performance of the deployed

IoT test-bed . But the system resource and requirements will change over time. So the

Autonomic Agent have to maintain a performance log and keep track the relationship
under which condition the system can perform the best. This is called the

”

'
phase of the Autonomic Agent, and can vary from minutes to weeks depending on the
con

' learning

guration. It also depends on the changes that the IoT system is facing. Customized
change can be imposed to the IoT platform to ' teach ' the system about different

scenarios and help him to de ne algorithm to look for best performance parameters. On

the other hand, it is also possible that the designer of the agent will pre-con gure the

' will depend on

the designer of the agent, and the preference given by the user or IoT administrator/
owner. The balance between two approaches is recommended.
During the analysis phase, the Autonomic Agent will develop a speci

best cases ' and load them in a pre-deployment scenario. The ' best case '

c mapping of

performance algorithm. How the ' best performance ' will be evaluated, totally depends

on the speci c provider or designer. In the current experimentation, a fastest response

to an IoT data access request has been considered as main goal. But it can range from