EdgeCoolingMode: An Agent Based Thermal Management Mechanism for DVFS Enabled Heterogeneous MPSoCs

Abstract

Thermal cycling as well as temperature gradient in time and space affects the lifetime reliability and performance of heterogeneous multiprocessor systems-on-chips (MPSoCs). Conventional temperature management techniques are not intelligent enough to cater for performance, energy efficiency as well as operating temperature of the system. In this paper we propose a lightweight novel thermal management mechanism in the form of intelligent software agent, which monitors and regulates the operating temperature of the CPU cores to improve reliability of the system. We validated our methodology on the Odroid-XU4 SoC and it has been successful to reduce the operating temperature by 6.32% while improving performance by 7.96% and reducing power consumption by 9.45% than the state-of-the-art. Index Terms-Lifetime reliability, multiprocessor systems-on-chip (MPSoCs), thermal management, thermal cycling, DVFS This work is supported by the UK Engineering and Physical Sciences Research Council EPSRC [EP/R02572X/1 and EP/P017487/1]. 1 DVFS helps to reduce the energy consumption by executing the workload over extra time at a lower voltage and frequency, which could be accounted for reduced power consumption.

P‐EdgeCoolingMode: an agent‐based performance aware thermal management unit for DVFS enabled heterogeneous MPSoCs

“…where l i : each frequency scaling level n : number of different frequency scaling level steps (6) Now, when the performance deadline is defined by the user in the P-EdgeCoolingModefor an executing application then from the operating frequency and achieved performance mapping deduced from Eq. 4 in the Learning module, the operating frequency, which is able to satisfy the defined performance deadline, is chosen from the mapping table instead of using the relationship deduced from Eq.…”

“…Whereas, Kirin 659 employs 4 ARM Cortex A-53 (big) CPUs and 4 ARM Cortex A-53 (LITTLE) CPUs and 2 ARM Mali-T830 GPU cores. Now a days most of the research and development in MPSoCs have been focused on developing algorithm to provide energy efficiency while catering for performance and reduced thermal gradient [3][4][5][6]. But in practical sense, majority of the published algorithms lack the flexibility for real world implementation because the usage and workload on similar devices by different users are different and hence, requires the resource management methodologies to be flexible to adapt over time.…”

“…In our previously published methodology, EdgeCoolingMode [6], we are able to reduce the operating temperature of the device by 6.32% (average) while improving the performance by 7.16% compared to Linux's Ondemand governor and reduce the operating * We chose the native option of the Streamcluster to mimic real-world data-mining algorithms, which are both compute intensive and memory intensive.…”

“…‡ Adjusting the clock speed of the CPU to use less energy consumption and reduce CPU temperature for improved reliability. 60 70 80 90 100 110 120 130 140 150 1520 1570 1620 1670 1720 1770 1820 1870 1920 1970 2020 1 13 25 37 49 61 73 85 97 109 121 133 145 157 169 181 193 205 217 229 241 253 265 277 289 301 313 325 337 349 361 373 385 [6] is aware of performance/throughput pro-actively while reducing thermal gradient then there is a possibility of reducing the operating temperature even more. In order to overcome the limitations of the existing approaches (as mentioned in Observation 1 and 2) towards addressing temperatureaware resource management mechanisms, which are capable of catering for both performance and reduced energy consumption, we propose a novel light-weight dynamic thermal management mechanism using intelligent software agents based on a fast heuristic approach.…”

TMAV: Temporal Motionless Analysis of Video using CNN in MPSoC

“…Although this study gave a novel approach on analyzing traffic using video cameras in low bandwidth network without the need to communicate the image frames over the WIFI network, the study also had energy consumption and device lifespan reliability issues (shown later in the section). Due to wide consumer adoption of mobile devices utilizing multi-processor system-on-a-chip (MPSoC) [26]- [28], which implements several different types of processing elements such as CPU/GPU on the platform, MPSoCs are perfect candidates for implementing computing resource demanding algorithms such as CNN based methodologies. When the CNN model proposed in the study [9] is implemented and trained on a MPSoC such as Odroid XU4 [29] (more on the hardware of Odroid XU4 in Sec.…”