Physical aware frequency selection for dynamic thermal management in multi-core systems

Abstract

In order to maintain performance per Watt in microprocessors, there is a shift towards the chip level multiprocessing paradigm. Microprocessor manufacturers are experimenting with tens of cores, forecasting the arrival of hundreds of cores per single processor die in the near future. With such large-scale integration and increasing power densities, thermal management continues to be a significant design effort to maintain performance and reliability in modern process technologies. In this paper, we present two mechanisms to perform frequency scaling as part of Dynamic Frequency and Voltage Scaling (DVFS) to assist Dynamic Thermal Management (DTM). Our frequency selection algorithms incorporate the physical interaction of the cores on a large-scale system onto the emergency intervention mechanisms for temperature reduction of the hotspot, while aiming to minimize the performance impact of frequency scaling on the core that is in thermal emergency. Our results show that our algorithm consistently succeeds in maximizing the operating frequency of the most critical core while successfully relieving the thermal emergency of the core. A comparison of our two alternative techniques reveals that our physical aware criticality-based algorithm results in 11.7% faster clock frequencies compared to our aggressive scaling algorithm. We also show that our technique is extremely fast and is suited for real time thermal management

Thermal Balancing Policy for Streaming Computing on Multiprocessor Architectures

“…Many recent works in computer architecture focus on power management and thermal control for multi-core and MPSoCs [9,5,4]. While reducing power density has the effect of reducing overall temperature, power-aware design does not directly imply that thermal gradients between different components are minimized or individual hot spots do not appear [13,5].…”

Predictive Dynamic Thermal and Power Management for Heterogeneous Mobile Platforms

“…To enhance thermal control and management further, [8,12] present task migration approach, where the tasks, which have been already scheduled to a particular resource are migrated to other resources in case of thermal violations. Researchers have even tried to implement different policies in conjunction with each other as proposed in [33]. Model predictive and optimal control theory have been recently employed for thermal management to achieve smooth control with minimal performance loss [45,50].…”

Online thermal control methods for multiprocessor systems

“…Dynamic power management [Benini and De Micheli 1998] was developed first for single processors and later extended to MPSoCs [Donald and Martonosi 2006;Mukherjee and Memik 2006]. Most methods use dynamic voltage and frequency scaling (DVFS) for processor power optimization and balancing [Mukherjee and Memik 2006].…”

“…Most methods use dynamic voltage and frequency scaling (DVFS) for processor power optimization and balancing [Mukherjee and Memik 2006]. The following sections describe how DVFS has been embedded in different thermal management policies to manage both the performance and the thermal profile of the MPSoC.…”

“…Many state-of-the-art thermal control policies manage power consumption via dynamic frequency and voltage scaling (DVFS) [Mukherjee and Memik 2006]. The structure of state-of-the-art DVFS-based thermal management systems is reported in the diagram of Figure 1.…”