Multicore Frequency Control via Core-Specific Load Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dynamic voltage frequency scaling (DVFS) techniques are not effective in multicore environments due to uniform frequency control across all cores, leading to inefficient power saving, as they do not account for varying load conditions and lack detailed core-specific power-saving policies.
Innovation Solution
A method that uses a scheduler to determine and adjust the frequency of each core based on its average load, comparing it to a threshold value, and setting the next frequency accordingly, with the option to turn off low-load cores, and allocating tasks to optimize power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If uniform frequency control is applied to all cores, then system simplicity is maintained, but power-saving efficiency deteriorates due to inability to account for varying load conditions
Solution Approach 1:
The patent divides the frequency control mechanism into core-specific segments. Each core has its own frequency control policy that is independently determined based on its load conditions. The scheduler evaluates each core separately and assigns appropriate frequency levels (e.g., high frequency for CPU-bound tasks, low frequency for I/O-bound tasks), enabling differentiated power management across cores while maintaining overall system coordination.
Solution Approach 2:
The patent implements local quality by tailoring frequency control characteristics to each core's specific workload requirements. Different cores can operate at different frequency levels simultaneously based on their individual task demands. This allows the system to apply appropriate frequency levels locally at each core rather than enforcing a uniform frequency across all cores, thereby reducing unnecessary power consumption.
2Productivity
If maximum frequency is used for all cores, then processing performance is maximized, but power consumption increases unnecessarily for low-load cores
Solution Approach 1:
The patent implements dynamic frequency adjustment where each core's frequency is continuously adapted based on real-time load conditions. The scheduler monitors task characteristics and dynamically changes frequency levels - assigning high frequencies to cores handling CPU-bound tasks requiring maximum performance, while assigning low frequencies to cores handling I/O-bound tasks with lower performance requirements. This dynamic adaptation ensures each core operates at the appropriate frequency level for its current workload.
Solution Approach 2:
The patent changes the frequency parameter of each core based on task type detection. When a core is assigned a CPU-bound task, the frequency parameter is increased to maximum or high levels to ensure processing performance. When a core is assigned an I/O-bound task, the frequency parameter is reduced to low levels to save power. This parameter adjustment is performed dynamically according to the actual workload characteristics of each core.
Data Source
AI summary
Provided is a method of dynamically controlling power in a multicore environment including a multicore processor which includes a plurality of cores and a scheduler. The method includes determining whether a management policy is set, collecting frequency change information used to change frequencies of the plurality of cores when it is determined that the management policy is set, calculating an average load of each core on a basis of the frequency change information, calculating an average frequency of each core according to the calculated average load of each core, comparing the average frequency of each core and a predetermined threshold value, and setting a next frequency of each core according to the comparison result.