Enhancing Energy Efficiency of Multimedia Applications in Heterogeneous Mobile Multi-Core Processors (2017)
Ref: Enhancing Energy Efficiency of Multimedia Applications in Heterogeneous Mobile Multi-Core Processors (2017), IEEE & Samsung Electronics
- saves system-wide (not just CPU) energy consumption by 8.9 percent
- Allocates multimedia applications to the small cores and non-multimedia applications to the big cores.
WAEAS
- improves 5-8% when the workload was high
WAEAS_An_optimization_scheme_of_EAS_scheduler_for_wearable_applications
Ref: An Optimization Scheme of EAS Scheduler for Wearable Applications (2021), TSINGHUA SCIENCE AND TECHNOLOGY
- BES-WALT Basic exponential smoothing-based WALT algorithm,
- Energy-aware CPU selection algorithm,
- Adjusts Sched Tune
- Latency-Sensitive Task: decrease searching speed to get minimum performance capacity and the lowest idle state
- Non-Latency-Sensitive Task: not choose a lower utilization, but “task packing strategy” ( on the little cores )
- Batch processing strategy,
- Adjusts select idle sibling()
- Record and centralize on fewer cores
- Cluster-based load balancing (overutilized)
- Reduces the meaningless task migration by local load balancing
- ( when having many aperiodic high-load applications )
Reinforcement Learning
Performance_Improvement_of_Linux_CPU_Scheduler_Using_Policy_Gradient_Reinforcement_Learning_for_Android_Smartphones.pdf
Learning EAS, LG Electronics
Ref: Performance Improvement of Linux CPU Scheduler Using Policy Gradient Reinforcement Learning for Android Smartphones (2020)
- improves power consumption by 2.8% - 7.8%
- Adjusts the TARGET_LOAD used to set the CPU frequency and the sched migration cost used as the task migration criteria
- Using Policy reinforcement learning dealing with workload or the ratio of sleep and running states changes.
KylinTune for Browser Engine
KylinTune_DQN-based_Energy-efficient_Model_for_Browser_in_Mobile_Devices.pdf)
Ref: DQN-based Energy-efficient Model for Browser in Mobile Devices
- EAS and Q-table / Deep-Q Network (DQN) for setting schedtune.boost in Browser Engine