With the rapid advancement of computer system architecture, more and more computersystems contain heterogeneous cores, and different cores might preserve differentlSAsand different computing capabilities.In addition, abundant newusage scenarios andcomplicated workloads further stimulate the demand for more computing power. Theabove facts call for a challenging research of exploiting the optimal utilization of systemresources to achieve extreme performance, in particular when taking specifc workloadcharacterization into consideration. Based on a hierarchical and quantitative analysis oftypical workloads in various application domains under production deployment, westudy the software-hardware co-optimization approach to effectively boosting theperformance of the workloads on a heterogeneous system, across clusters inside adatacenter, or within a decentralized environment, The quantitative analysis also shedslight on well designing the domain specific architectures for particular applications.Furthermore, we also explore the possibilities of developing more effective tools onworkload characterization, performance profling, bottleneck localization and programoptimization. On talent cultivation perspective, we expect our students could grasp solidcapabilities and skills on system architecture and performance engineering.
