Energy-Aware Design of Computing Systems
The seed of the course is the presence of the energy/power/thermal showstoppers in the design of computing systems. The negative impact of such problems is actually manifest at any abstraction level of the design of the applications, from the hw up to the cloud-class application. Many attempts have been made to tackle such problems, but most of them focused on a specific aspect of the overall picture. For example energy simulation of software execution or modeling of the total cost of ownership of big data centers including cooling and maintenance, as well as dynamic resource adaptation. The advent of the multi/many-core era, is enabling a new perspective in the design of systems but exacerbated the need of a vertical and unified energy aware design of applications and architectures, due to the so called presence of “dark silicon”, i.e. of an even increasing portion of the system which has to be kept switched off to prevent possible systems failures, with a poor resource utilization and increasing of the costs. The impossibility of achieving a simultaneous power-on of cores due to thermal and power density constraints, based on the technological data from ITRS and Intel, has an exponential growth with each. For instance, at 8nm which is far just a coming decade, more than 50% of the chip area will be dark (unusable). Hence, architecture and system design communities have started coping with such problems from many perspectives, such as intelligent power management, heterogeneous architectures, power aware design of software and design of specific accelerators. The goal of the course is to collect the state of the art in this field, to be compared with the available commercial or popular solutions, with the goal to provide a set of design methodologies usable at different abstraction layers. The knowledge of the overall picture should make possible to the designer to consider and focus on the “real problem” of designing systems for the next generation of applications, account also for the inter-layer possible synergies. The presence of a project work is a mean to familiarize the students not only with real problems but also with realistic solutions, from a critical standpoint.
Syllabus
- Introduction and motivation
- Green Computing in Data centres
- Energy/Power-Aware Compiler techniques
- Power & Security: Side Channel Attacks and countermeasures
- Power, energy and thermal viewpoints on architectures
- Runtime Resource Management, monitors and knobs
- Wireless Sensor Networks and Cyber-Physical Systems
- Energy estimation in embedded software development
