Multivariable Control for Data Center Power and Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern data centers face challenges in managing power consumption and latency due to their large number of computing servers, leading to inefficiencies and high electricity costs, especially during non-peak times when resources are underutilized.
Innovation Solution
A multivariable control system that identifies demand for computing resources and adjusts the operating states or modes of servers to minimize power consumption while maintaining adequate response times, using techniques like Robust Multivariable Predictive Control (RMPCT) to optimize server states and modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If data centers are designed with over-capacity to maintain acceptable latency during peak times, then latency performance is improved, but power consumption increases and resource utilization decreases during non-peak times
Solution Approach 1:
The patent applies dynamics by transitioning computing devices between different operating states (active, standby, hibernation) based on real-time demand conditions. The multivariable controller dynamically adjusts the number of devices in each state, allowing the system to adapt its power consumption and latency characteristics to match varying workload demands, thereby resolving the contradiction between maintaining low latency and reducing power consumption.
2Use of energy by stationary object
If computing devices are placed in power-saving states to reduce power consumption, then power usage is improved, but response time to user demands worsens
Solution Approach 1:
The patent applies preliminary action by maintaining a pool of computing devices in standby states that are pre-positioned and ready to quickly respond to demand spikes. The multivariable controller predicts future demand based on historical patterns and proactively activates devices before peak periods occur, ensuring rapid response times while minimizing the duration devices spend in high-power states.
Solution Approach 2:
The system implements feedback through the multivariable controller that continuously monitors actual response times and power consumption, adjusting the distribution of devices across operating states in real-time. This closed-loop control ensures that response time requirements are met while optimizing power consumption by reducing the number of devices in active states when demand is low.
3Productivity
If the number of computing devices is increased to handle peak demand, then productivity is improved, but device complexity and operational management become more difficult
Solution Approach 1:
The patent applies universality by designing computing devices with multiple operating states that can serve different functional requirements. The same physical device can function as an active processing unit, a standby resource, or a hibernating unit depending on system needs, thereby reducing the total number of devices required and simplifying operational management while maintaining the capacity to handle peak demand.
Data Source
AI summary
A method includes identifying demand for computing resources provided by multiple computing devices and identifying operating states or modes for the computing devices based on the demand using a multivariable controller. The multivariable controller is configured to determine how to alter multiple manipulated variables in order to create changes to multiple controlled variables. The multiple manipulated variables include the operating states or modes of the computing devices, and the multiple controlled variables include a power consumption of the computing devices and a response time of the computing devices. Each of the computing devices could include one or more processing units, and each of the computing devices or processing units could be configured to selectively operate in one of the operating states or modes. The method could also include generating a profile identifying a number of computing devices or processing units to operate in each of the operating states or modes.