Next: References Up: Grid Computing : An Previous: Abstract Contents
The upper tier, the users, will generate tasks with defined parameters for execution such as hardware / software requirements and will then be sent off to the middle tier to locate the optimum resource centre. It will be possible for the model to automatically generate tasks at random intervals and perform basic statistical analysis on average completion time, etc.
The middle tier will be controlled by a number of agent ‘brokers’ who will attempt through their local knowledge of the world to pair any given task with the most appropriate resource centre. It should be noted that due to the restrictions of the tools the agents modelled will be linear instead of concurrent, using pre-defined dependencies to govern the flow data through this layer.
Resource centres (data processing nodes) within the lower tier will be assigned a hardware / software state as well as a resource usage level (queuing time). It will be this state which will govern the overall probability of a task being gained by an individual resource.
The simplistic model which will be generated will assume a benevolent system where by any state information filtered up to the upper tier will be both current and correct. Whilst this is often not the case in real world simulations, inserting delays and agent bias into the system will likely cause an unstable and unrealistic simulation.
The model will be completed in such a way as to provide an educational insight into the relatively new field of grid computing to users not already familiar with the structure and processes. As such it will have the ability to run in a simulation mode whereby tasks are automatically generated and assigned. Users can then follow the actions within the system to understand the brokering and negotiations involved in the task designation.