In this work, we propose a method of comprehensive design space exploration for computer clusters built with mass-produced, off-the-shelf components, paving the way to a future CAD system with decision support capabilities. We formulate the design task as a discrete combinatorial optimisation problem, with the non-linear objective function being the ratio of total cost of ownership to real-life performance. Various constraints can be imposed on technical and economic characteristics of the computer cluster. Obtaining the value of the objective function is only possible after several consecutive stages of design process are completed. We provide a prototype CAD tool that implements these stages. We also propose heuristics to deal with combinatorial explosion at various stages.
«
In this work, we propose a method of comprehensive design space exploration for computer clusters built with mass-produced, off-the-shelf components, paving the way to a future CAD system with decision support capabilities. We formulate the design task as a discrete combinatorial optimisation problem, with the non-linear objective function being the ratio of total cost of ownership to real-life performance. Various constraints can be imposed on technical and economic characteristics of the compu...
»