Nowadays, peak computational performance in many problem domains is achieved through the use of specialized hardware such as graphics processing units (GPUs) and field-programmable gate arrays (FPGAs). The goal of this project is to allow CityMoS to make use of modern heterogeneous hardware to accelerate city-scale simulation-based evaluations.
Approach and Methodology
Applying heterogeneous hardware for accelerating simulations poses two challenges: first, hardware supporting highly parallel processing such as GPUs and FPGAs exhibits highest performance for particular types of computations only. Second, when partitioning the simulation workload among processing elements, the required communication may outweigh the gain in processing power. Thus, in this project we will develop methods to characterize the computations performed in different parts of a simulator and to profile communication demands between different parts of a simulator. Using the gained information, the simulation can be partitioned into sub-tasks and assigned to the processing elements of the heterogeneous platform.
Outcomes and Benefit
The outcome of this project will be methods for profiling and partitioning CityMoS to make use of heterogeneous hardware as well as implementations of individual simulation tasks runnable on GPUs or FPGAs. Through the improved utilization of the available hardware, the time required to perform city-scale evaluations using CityMoS will be substantially reduced.