RewardProfiler: A Reward Based Design Space Profiler on DVFS Enabled MPSoCs

Abstract

Resource mapping on a heterogeneous multiprocessor system-on-chip (MPSoC) imposes enormous challenges such as identifying important design points for appropriate resource mapping for improved efficiency or performance, time consumption of exploring all the important design points for each profiled applications, etc. Moreover, incorporating a profiler into integrated development environments (IDEs) in order to achieve more detailed and accurate profiling information on the application being targeted during runtime such that improved efficiency or performance while executing the application is achieved, the runtime resource management decision to achieve such improved "reward" has to be utilized in a certain way. In this paper, we propose a hybrid approach of resource mapping technique on DVFS enabled MPSoC, which is suitable for IDE integration due to the reduced design points in our methodology resulting in significant reduction in profiling time. We coined our approach as "RewardProfiler" (a Reward based design space Profiler), which is well capable of reducing the design space exploration without losing most of the important design points based on our heuristic approach. In our strategy, an application has to be mapped onto the available resources in such a way so that the "reward" obtained can be maximized. Our approach can also be utilized to maximize multiple "rewards" (Multivariate Reward Maximization) while executing an application. Implementation of our RewardProfiler on the Exynos 5422 MPSoC reveals the efficacy of our proposed approach under various experimental test cases and has a potential of saving 170x more time in profiling for our chosen MPSoC compared to the state-of-the-art methodologies.

Dynamic Energy and Thermal Management of Multi-core Mobile Platforms: A Survey

“…However, majority of the studies are focused on many-core (more than 16 cores) general purpose processors and Network-onchips (NoCs) on contrary to multi-core mobile platforms. Techniques in [9], [39], [40] are solely focused on optimizing thermal behaviour during runtime on mobile platforms. Note that this section covers approaches considering only thermal management but not both thermal and energy management that are provided in the next section.…”

“…The efficacy of the technique is proved on Samsung Exynos 5422 MPSoC. In another work [9], Dey et al presented a dynamic thermal management technique where design space exploration is used to first reduce the number of possible frequencies to only four and then selecting the most appropriate frequency to meet the desired reward, which is the thermal constraint for an example.…”

“…For decades, several research and implementation works have focused on optimizing energy at circuit, architecture and system levels. According to [9] there are five popular methods and/or combination of them leading to energy reduction in the system, which includes: 1) Dynamic Power Management (DPM) allows idle processing elements (PEs) or other idle components of the system to be suspended if required in order to reduce energy consumption. 2) Dynamic Voltage Frequency Scaling (DVFS) allows processors to operate at variable voltage-frequency (v-f) levels.…”

ThermalAttackNet: Are CNNs Making It Easy To Perform Temperature Side-Channel Attack In Mobile Edge Devices?

“…Most of these mobile devices come equipped with heterogeneous multi-processors systemson-chip (MPSoC), which consists of multiple heterogeneous processors on a single chip, capable of processing different types of applications to cater for performance and energyefficiency of the executing applications. Due to an increase in the usage of MPSoCs [2], [8]- [13] in mobile edge devices and a rise in studies on thermal side channel attacks [6], [14], [15], it is crucial that side channel attacks in such platform should be addressed with utmost importance [15].…”

FruitVegCNN: Power- and Memory-Efficient Classification of Fruits & Vegetables Using CNN in Mobile MPSoC

“…On the other hand, battery operated mobile devices are becoming more affordable due to the improvement in chip manufacturing technology and also come equipped with multiple processing elements to cater for performance requirement of the executing applications [10], [11]. Such a mobile platform, utilizing multi-processor system-on-a-chip (MPSoC) [10], [11], [12], [13], [14], [15], [16], [17], implements different types of processing elements such as CPU and GPU, capable of computing demanding computer vision applications on the device. Given the importance of adopting CNN based approaches in agricultural industry and the increase in popularity of using mobile devices, it is very important to study, design and implement CNN models for fruits and vegetables'…”