Accelerator Power and Cooling Control for Variable Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for accelerator-equipped servers fail to achieve high electricity efficiency and responsiveness when input data varies in time, leading to redundant computing power and overconsumption, while also lacking support for various accelerators.
Innovation Solution
An accelerator state control device that predicts input data variation, determines optimal computing power and cooling settings, and applies these settings to the accelerator and cooling mechanism to maintain efficiency and responsiveness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the accelerator computing power is increased to handle peak traffic, then responsiveness is improved, but electricity consumption increases during low-traffic periods
Solution Approach 1:
The patent applies dynamics by making the accelerator computing power adjustable rather than fixed. The system dynamically changes the operating frequency of the accelerator according to real-time traffic conditions, increasing frequency during high-traffic periods to maintain responsiveness and decreasing frequency during low-traffic periods to reduce electricity consumption. This is achieved through a frequency adjustment module that receives traffic information and correspondingly adjusts the accelerator's operating frequency.
Solution Approach 2:
The patent applies parameter changes by modifying the operating frequency parameter of the accelerator based on traffic conditions. The system changes the frequency parameter dynamically - increasing it when traffic is heavy to improve processing speed, and decreasing it when traffic is light to reduce power consumption. This parameter adjustment directly addresses the contradiction between maintaining speed and reducing energy use.
2Use of energy by moving object
If the accelerator computing power is reduced to save electricity during low traffic, then electricity efficiency is improved, but responsiveness deteriorates during high traffic
Solution Approach 1:
The patent applies feedback by implementing a closed-loop control system where traffic information is continuously monitored and fed back to adjust accelerator frequency. The feedback module receives real-time traffic data, compares it with threshold values, and automatically adjusts the accelerator's operating frequency accordingly. This ensures the system maintains optimal responsiveness during high traffic while achieving energy efficiency during low traffic periods.
Solution Approach 2:
The system dynamically adapts the accelerator's computing power to match actual traffic demands. During low-traffic periods, the frequency is reduced to improve electricity efficiency. During high-traffic periods, the frequency is increased to maintain processing speed and responsiveness. This dynamic adjustment resolves the contradiction between energy efficiency and speed performance.
3Reliability
If the cooling mechanism operates at high capacity to cool the accelerator, then reliability is improved, but electricity consumption increases
Solution Approach 1:
The patent applies dynamics to the cooling mechanism by making its capacity adjustable rather than fixed. The cooling system dynamically changes its operating capacity according to the accelerator's actual heat generation, which correlates with traffic load. During high-traffic periods, the cooling capacity is increased to maintain reliability. During low-traffic periods, the cooling capacity is reduced to decrease electricity consumption.
Solution Approach 2:
The system implements periodic adjustment of cooling capacity based on traffic patterns. The cooling mechanism operates at different capacities during different time periods - high capacity during peak traffic periods when reliability is critical, and low capacity during off-peak periods when energy savings are prioritized. This periodic action resolves the contradiction between reliability and energy consumption.
4Use of energy by moving object
If the cooling capacity is reduced to save electricity during low traffic, then electricity efficiency is improved, but reliability deteriorates during high traffic
Solution Approach 1:
The patent applies feedback by implementing real-time monitoring of traffic conditions and automatic adjustment of cooling capacity. When traffic increases and cooling demand rises, the feedback mechanism detects this change and increases cooling capacity to maintain reliability. When traffic decreases, the system reduces cooling capacity to improve electricity efficiency. This feedback loop ensures reliability is maintained during high traffic while achieving energy efficiency during low traffic.
Solution Approach 2:
The system changes the cooling capacity parameter dynamically based on traffic conditions. During low-traffic periods, the cooling capacity parameter is reduced to improve electricity efficiency. During high-traffic periods, the parameter is increased to ensure operational reliability. This parameter adjustment resolves the contradiction between energy consumption and reliability.
5Productivity
If the accelerator configuration is optimized for specific accelerator types, then productivity is improved, but adaptability deteriorates
Solution Approach 1:
The patent applies universality by designing a frequency adjustment mechanism that works across multiple accelerator types (FPGA, GPU, ASIC). The system maintains a library of frequency adjustment patterns applicable to different accelerator architectures, allowing the same control logic to optimize productivity for various accelerator types while maintaining adaptability through type-specific parameter configurations.
Solution Approach 2:
The system optimizes productivity by adjusting frequency parameters specific to each accelerator type while maintaining a universal control framework. For each accelerator type (FPGA, GPU, ASIC), the system applies type-appropriate frequency adjustments that maximize computing productivity, while the overall architecture remains adaptable to support multiple accelerator technologies through parameterized configurations.
Data Source
AI summary
An accelerator state control device includes: an input amount acquisition/prediction unit that predicts amount of processing to be offloaded to an accelerator and outputs a prediction result as a traffic and a variation range of the traffic; a computing power setup recording unit that maintains information on a type and a model of the accelerator and setup information tailored to performance, as a list, and retrieves information from the list in response to an inquiry and responds to the inquiry; an ACC computing power/consumed electricity setup determination unit that determines setup information on computing power and varying time of the accelerator, based on the traffic and the variation range as well as the list; and an ACC computing power/consumed electricity setting unit that applies the setup information to the accelerator, based on the setup information for the accelerator determined by the ACC computing power/consumed electricity setup determination unit.