Section: Computer Science and Engineering
Tutor: BONARINI ANDREA Major Research topic
:Multi-core resource management: a horizontal perspective
Advisor: FORNACIARI WILLIAMAbstract:
Modern computing systems strive to provide ever increasing performance levels despite increasingly strict system-wide optimization objectives. This is a vast problem that spans over a wide variety of architectures: due to the wild technological development caused by the spread of devices such as Smartphones, high-end embedded systems are quickly closing the gap with desktop computers; similarly, high performance and cloud-based systems are scaling up towards exa-scale to serve increasingly demanding workloads.
Indeed, this technological trend poses several, nontrivial problems: embedded systems are usually subject to thermal and energy constraints in order to maximize battery life, to minimize faults and, at least in the case of hand-held devices, to provide a comfortable user experience (users want to hold a long lasting, ever charged and cold device), while bigger systems are typically subject to thermal and power constraints in order to minimize supplying/cooling costs and, again, to prevent faults. On the other hand, users do not care about system optimization objectives: they just want their applications to comply with some Quality of Service requirement.
This problem does not have a simple solution, because system and user goals are orthogonal and increasingly demanding; however, it can be addressed by employing resource managers, which are software layers that act as brokers between computing systems and applications. Resource managers decide which and how many resources will be allocated to each application so that, whenever it is possible, both system-wide and user goals are complied with.
My work explores the problem of resource management from a horizontal perspective. That is, I analyze the problem of CPU resource management spanning from high-end embedded to High Performance Computing systems. For each of those architectures, I try to understand what is yet missing to obtain an optimal resource management and how we can fill some of those gaps.
Advisor: William Fornaciari