Collaborative Research: CCRI: Planning-C: Accelerated Infrastructure for Simulating Future Systems

Software-based simulation is the backbone of computer architecture research and development. The project novelty is the identification of the following problems with these methodologies: (1) The use of on-premise servers for running simulations suffers from the high operation and embodied energy costs. Although simulation performance has always been a primary engineering focus of software-based simulators, energy consumption and environmental impacts of detailed hardware simulation have mainly been ignored. (2) Detailed software-based simulations have different run-time characteristics than baremetal applications.

Such differences make the general-purpose servers with default server configurations extremely inefficient for running detailed simulations. (3) Lastly, full-system simulation environments have steep learning curves that further steepen as the hardware and software stacks to be simulated become more complex. The project’s impacts are to bring the CISE community's attention to the energy waste at the onpremise simulation clusters and provide a foundation for minimizing the energy and carbon footprint of future hardware evaluation.

The investigators’ goal is to create a cloud-based community infrastructure for running simulations. This community infrastructure eliminates the need for funding agencies such as NSF to invest in on-premise servers and significantly reduce the end-to-end carbon footprint of simulations. More specifically, investigators extensively analyze the source code and run-time of the popular architectural simulators to understand their characteristics as a software program that needs to be efficiently executed on real hardware.

The investigators use the profiling results to tune system parameters and provide a highly optimized environment for running architectural simulations. They enhance the proposed cloud infrastructure with hardware accelerators that further improve popular architectural simulators' energy efficiency and performance. The infrastructure follows a marketplace model for sharing and using simulation models and setups developed by other researchers to reduce the learning curve of detailed simulations.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.