Multicore computer systems are adopting more cores to meet the increasing performance requirement. Network topology to make communication in such a multicore/manycore system (like Intel 80-Core Research Chip and IBM Roadrunner) is extremely important to achieve high performance by supplying low energy. Designing and analysing such a gigantic system is very complicated and it costs up to millions of dollars. Performance modelling technique is attracting researches from many organizations and different disciplines to conduct research on multicore/manycorenetworked systems in an inexpensive way. However, there is no standard in performance modelling of such complex systems. This paper proposes a methodology of developing conceptual models of multicore/manycorenetworked systems for performance analysis. Important features of this methodology include data acquisition to improve accuracy by capturing the important details about the system under consideration, system-to-model abstraction to make the modelling manageable without eliminating any characteristics of the target system, and model validation to make sure that the model meets the requirements with an optimal choice of the hardware and software. Networked systems with up to 64 cores are modelled; according to the experimental results, the proposed performance modelling methodology is very promising
展开▼
