Tendra lugar este jueves 4 de mayo, de 12:00 a 13:00 en el Aula A.05 del Edif. Ada Byron
Summary:
Hardware performance monitoring counters (PMCs) have proven effective in characterizing application performance. A large body of work has demonstrated that several components of the operating system (OS), such as the scheduler, can perform effective runtime optimizations in multicore systems by leveraging performance-counter data. While existing tools greatly simplify the collection of PMC data from user space, they do not provide an architecture-agnostic mechanism that is capable of exposing high-level PMC metrics to the OS. Thus, the implementation of OS-level PMC-driven optimization schemes is typically tied to specific processor models.
In this talk I will present PMCTrack, an open-source tool for the Linux kernel that seamlessly enables the system software to access PMC data in an architecture-independent fashion, and also provides other insightful monitoring information available in modern processors, such as cache occupancy or energy consumption. Despite being an OS-oriented tool, PMCTrack still allows the gathering of monitoring data from user space, making it possible for users and developers to perform offline analysis in various ways.
Short bio
Juan Carlos Saez received his Ph.D. in computer science in 2011 from the Complutense University of Madrid (UCM), where he also obtained the Extraordinary Doctorate Award. He is now an Associate Professor in the Department of Computer Architecture at UCM. Since 2013, he has served as the UCM Campus Representative of USENIX, the Advanced Computing Systems international association. In the last few years, he has been teaching different courses related to Operating Systems and Computer Architecture. His research interests include energy-aware computing and improving the interaction between the system software and hardware for emerging architectures. His recent research activities focus on OS scheduling on heterogeneous multicore processors, exploring new techniques to deliver better performance per watt, and quality of service on these systems.
