Multiprocessor Energy Saving via Sleep Task Injection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current power management policies in multiprocessor systems, such as timeout and predictive policies, often result in wasted energy and performance penalties due to incorrect estimates and inefficient state transitions between low-power states, particularly in systems with non-exponential inter-arrival and execution times.
Innovation Solution
A multiprocessor system that includes a counting, measuring, and calculating unit to determine the generating rate and time length of sleep tasks based on an acceptable delay, which are then injected into the task traffic to switch off processors, thereby reducing power consumption and controlling average response times without increasing response times of original tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If timeout policy is used to manage power states, then power consumption is reduced, but energy is wasted while waiting for timeout events and response time increases
Solution Approach 1:
The system performs preliminary actions by generating sleep tasks in advance based on predicted idle periods. The CMC unit analyzes task arrival patterns and proactively creates sleep tasks that will be injected into the task queue, allowing processors to transition to low-power states before actual idle periods occur, rather than waiting for timeout events
Solution Approach 2:
The system implements feedback mechanisms where the CMC unit continuously monitors task arrival rates and processor utilization, adjusting the generation rate and time length of sleep tasks dynamically. This feedback loop allows the system to optimize the balance between power savings and response time based on actual system conditions
2Loss of energy
If predictive policies force transition to low power state, then energy saving is achieved, but performance deteriorates due to incorrect estimates
Solution Approach 1:
The system dynamically adjusts sleep task parameters based on real-time system conditions. The CMC unit continuously updates the generation rate and time length of sleep tasks according to observed task arrival patterns and processor utilization, making the power management policy adaptive rather than static. This allows the system to optimize energy savings while maintaining performance under varying workloads
Solution Approach 2:
The system changes parameters of sleep tasks (generation rate, time length) based on system state. By adjusting these parameters dynamically, the system can achieve better energy savings without causing performance degradation from incorrect predictions, as the parameters are adapted to actual workload conditions
3Adaptability or versatility
If frequent state transitions are performed, then power management responsiveness is improved, but energy is wasted on transitions and performance deteriorates
Solution Approach 1:
The system uses periodic action by implementing sleep tasks with specific time lengths that control the duration of low-power states. Instead of frequent transitions, the system establishes longer-duration sleep periods based on predicted idle times, reducing the number of transitions while maintaining effective power management. The CMC unit determines optimal time lengths to minimize transition overhead
Data Source
AI summary
A multiprocessor system comprises: a plurality of processors; a counting, measuring and calculating (CMC) unit that determines a generating rate of sleep tasks and a time length of each of the sleep tasks based on an acceptable delay; a sleep task generator that generates the sleep tasks with the time length at the generating rate, and injects the generated sleep tasks into a traffic for original tasks; and a scheduler that assigns both the original tasks and the sleep tasks in the traffic to the plurality of processors, wherein each of the sleep tasks switches off one of the plurality of processors, on which the sleep task is assigned.