Learning Pareto-Frontier Resource Management Policies for Heterogeneous SoCs: An Information-Theoretic Approach

  title={Learning Pareto-Frontier Resource Management Policies for Heterogeneous SoCs: An Information-Theoretic Approach},
  author={Aryan Deshwal and Syrine Belakaria and Ganapati Bhat and Janardhan Rao Doppa and Partha Pratim Pande},
  journal={2021 58th ACM/IEEE Design Automation Conference (DAC)},
Mobile system-on-chips (SoCs) are growing in their complexity and heterogeneity (e.g., Arm’s Big-Little architecture) to meet the needs of emerging applications, including games and artificial intelligence. This makes it very challenging to optimally manage the resources (e.g., controlling the number and frequency of different types of cores) at runtime to meet the desired trade-offs among multiple objectives such as performance and energy. This paper proposes a novel information-theoretic… 

Figures and Tables from this paper

Adaptive Experimental Design for Optimizing Combinatorial Structures

The key challenges in solving combinatorial spaces problems in the framework of Bayesian optimization (BO) and the progress over the last five years in addressing these challenges are described.



Dynamic Resource Management of Heterogeneous Mobile Platforms via Imitation Learning

The Oracle policies enable us to design low-overhead power management policies that achieve near-optimal performance matching the Oracle, and present efficient approaches for constructing an Oracle policy to optimize different objective functions, such as energy and performance per Watt.

Imitation Learning for Dynamic VFI Control in Large-Scale Manycore Systems

This work proposes the first architecture-independent IL-based methodology for dynamic VFI (DVFI) control in manycore systems and demonstrates that IL is able to obtain higher quality policies than RL with significantly less computation time and hardware area overheads.

HiLITE: Hierarchical and Lightweight Imitation Learning for Power Management of Embedded SoCs

This work proposes HiLITE, a hierarchical imitation learning framework that maximizes the energy efficiency while satisfying soft real-time constraints on embedded SoCs and shows that the trained policies not only achieve high accuracy, but also have negligible prediction time overhead and small memory footprint.

An Energy-aware Online Learning Framework for Resource Management in Heterogeneous Platforms

This work proposes an online imitation learning approach to construct an offline policy and adapt it online to new applications to optimize a given metric (e.g., energy) and demonstrates its effectiveness on a commercial mobile platform with 16 diverse benchmarks.

Evolutionary Algorithms for Multiobjective Optimization

A multi-objective evolutionary algorithm with aggregate surrogate model, its newer version, which also uses a surrogate model for the pre-selection of individuals and a framework which combines surrogate modeling with meta-learning is presented.

Handbook of Differential Entropy

Handbook of Differential Entropy provides a comprehensive introduction to the subject for researchers and students in information theory and describes common estimators of parametric and nonparametric differential entropy as well as properties of the estimators.

A closer look at drawbacks of minimizing weighted sums of objectives for Pareto set generation in multicriteria optimization problems

A standard technique for generating the Pareto set in multicriteria optimization problems is to minimize (convex) weighted sums of the different objectives for various different settings of the

DyPO: Dynamic Pareto-Optimal Configuration Selection for Heterogeneous MpSoCs

This paper proposes a novel methodology that can find the Pareto-optimal configurations at runtime as a function of the workload, and uses an extensive offline characterization to find classifiers that map performance counters to optimal configurations.

Inter-Cluster Thread-to-Core Mapping and DVFS on Heterogeneous Multi-Cores

A run-time management approach that first selects thread-to-core mapping based on the performance requirements and resource availability and applies online adaptation by adjusting the voltage-frequency levels to achieve energy optimization, without trading-off application performance is proposed.

ML-Gov: a machine learning enhanced integrated CPU-GPU DVFS governor for mobile gaming

A machine learning enhanced integratedCPU-GPU governor that builds tree-based piecewise linear models offline, and deploys these models for online estimation into an integrated CPU-GPU Dynamic Voltage Frequency Scaling (DVFS) governor, which achieves significant energy efficiency gains.