Distributed System Node Power Management via Recovery Time Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing distributed systems face challenges in reducing power consumption while maintaining processing performance, as node resumption times cannot be hidden in unpredictable workloads, leading to decreased performance when nodes are put in idle states.
Innovation Solution
A distributed system with management nodes that selectively restore ordinary nodes to active states based on recovery time, allowing for efficient power management by prioritizing nodes with shorter recovery times, thereby reducing overall power consumption without compromising performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If nodes are put in idle states to reduce power consumption, then power consumption is reduced, but processing performance deteriorates when load increases due to node resumption time
Solution Approach 1:
The management node performs preliminary actions by predicting future job arrivals and proactively restoring nodes from idle states before they are actually needed. This allows the system to reduce power consumption by keeping nodes idle during low-load periods while ensuring nodes are ready before load increases, thus avoiding performance deterioration.
Solution Approach 2:
The system dynamically adjusts node states based on predicted workload. The management node uses prediction information to flexibly transition nodes between active and idle states, optimizing the balance between power consumption and processing performance according to anticipated system conditions.
2Productivity
If nodes are restored to active states quickly to maintain performance, then processing performance is maintained, but power consumption increases
Solution Approach 1:
Instead of restoring nodes immediately when load increases, the system performs preliminary restoration based on predicted job arrivals. This allows nodes to be restored in advance during periods when full performance is not yet required, spreading out the power consumption over time rather than having sudden spikes.
Solution Approach 2:
The system maintains continuous useful action by keeping prediction functionality always active and nodes in transitional states. Rather than abrupt transitions between full power and idle states, the system continuously monitors predictions and gradually adjusts node states, maintaining optimal performance while managing power consumption smoothly.
Data Source
AI summary
A distributed system includes: a plurality of ordinary nodes provided with reduced-power states having different times of recovery to a normal operating state; and a management node for assigning a job to an ordinary node for carrying out the job. The management node has: node select means for selecting an ordinary node from ordinary nodes each put in one of the reduced-power states, assigning a job to the selected ordinary node and driving the selected ordinary node to carry out the assigned job; and node control means for executing control to restore an ordinary node selected by the node select means to the normal operating state. The node select means selects an ordinary node from the ordinary nodes each put in one of the reduced-power states having different times of recovery to the normal operating state in accordance with an ordinary-node order starting with an ordinary node existing in a reduced-power state and having a short time of recovery to the normal operating state.