Network Interface Wake on Compute Power Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data centers with multi-device platforms, a significant portion of hardware components remain unused but powered on, leading to increased power consumption and energy wastage, as existing power management technologies like DVFS and C-states do not effectively account for the power usage of all components, including accelerators, storage, and network devices.
Innovation Solution
The Wake on Compute (WoC) technology equips network interface devices with circuitry to respond to instructions while in a reduced power state, allowing for dynamic power management of components such as GPUs, CPUs, and accelerators, enabling them to be powered off or on as needed, thereby reducing overall energy consumption and operational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If hardware components are kept powered on to ensure immediate availability, then system responsiveness is improved, but power consumption increases
Solution Approach 1:
The system dynamically adjusts the power state of hardware components based on workload demands. Components transition between active and reduced-power states, with the network interface device enabling selective wake-up of specific components only when needed, optimizing the balance between responsiveness and energy efficiency
Solution Approach 2:
The network interface device maintains a low-power listening state continuously to detect wake-up packets, enabling rapid system activation without requiring all components to remain fully powered. This preliminary low-power state allows the system to respond quickly to workloads while minimizing baseline power consumption
2Loss of energy
If DVFS and C-states are used to reduce power consumption, then energy efficiency is improved, but they fail to account for power usage of all components including accelerators and storage
Solution Approach 1:
The network interface device serves as a universal power management coordinator that can control multiple different types of hardware components (GPUs, CPUs, accelerators, storage devices, network devices) through a single interface. It sends wake-up packets to various components regardless of their specific type, providing comprehensive power management across the entire system
Solution Approach 2:
The network interface device acts as an intermediary between the network and various hardware components. It receives wake-up packets from the network and translates them into component-specific activation signals, enabling power management of components that would otherwise not respond to standard power management protocols
3Loss of energy
If unused hardware components are powered off to save energy, then power consumption is reduced, but system availability and responsiveness deteriorate
Solution Approach 1:
The system segments power management control at the component level rather than system-wide. Individual hardware components can be independently powered off or awakened based on specific workload requirements. The network interface device can selectively activate only the necessary components for a given task, maintaining system availability for required functions while powering down unused components
Data Source
AI summary
Examples described herein relate to a network interface device that includes circuitry to perform switching and perform a received command in one or more packets while at least one of the at least one compute device is in a reduced power state, wherein the command is associated with operation of the at least one of the at least one compute device that is in a reduced power state. In some examples, the network interface device is able to control power available to at least one compute device.