The work will offer a step towards the feasibility of this notion by developing node architecture with a view to economize the execution of grand challenge applications. This is initiated by refurbishing the role of a single node from being specific to one particular application, to supporting simultaneous execution of more than one in a cluster environment by making it handle traces of multiple applications, leading to the design of a cost-effective supercomputer.
To the best of our knowledge, there is no simulator which offers design scalability to help steer the designer through the design space for supercomputer nod architectures. We will present a design space based on the proposed node architecture model with a simulator that is being developed. While the designer is initially expected to make coarse design choices from the design space governed by the model, the simulator will guide him/her through the design space to arrive at optimized design solutions to develop node architecture for the targeted cluster environment, taking into consideration all the constraints imposed by technology, performance requirements and performance-power tradeoffs. The potential of such node architecture can only be fully exploited with appropriate cluster architecture.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment