Resource Manager Proxy for Distributed Power Configuration

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

High-performance computing (HPC) systems face challenges in efficiently managing energy consumption due to the tension between global and local power configuration management, leading to suboptimal energy efficiency and performance, as existing Resource Managers struggle to make timely decisions across large numbers of computers.

Innovation Solution

Implementing a Resource Manager Proxy in each computer to monitor execution states and communicate with a centralized Resource Manager, enabling fast and responsive power configuration changes based on the application's state, allowing for fine-grained control of power configurations across the HPC cluster.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Stability of the object's composition

If centralized Resource Manager makes power configuration decisions for all computers, then energy management coordination is improved, but response time deteriorates due to communication and processing delays

Engineering Contradiction:

Improveenergy management coordinationVSAvoidresponse time

Core Design Contradiction:

Stability of the object's compositionVSLoss of time

Solution Approach 1:

The system segments power configuration management into two parts: a centralized Resource Manager that makes decisions based on global application state, and local Resource Manager Proxies that execute decisions immediately on each computer. This segmentation allows coordinated energy management at the global level while achieving fast local response times without communication delays.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The Resource Manager Proxy acts as an intermediary between the centralized Resource Manager and the local computer hardware. The Proxy receives power configuration decisions from the Resource Manager and immediately implements them locally, eliminating the time delay that would occur if the Resource Manager directly controlled each computer. The Proxy mediates between global coordination needs and local response requirements.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Speed

If local power management is implemented on each computer, then response speed is improved, but system-wide energy optimization deteriorates due to lack of global coordination

Engineering Contradiction:

Improveresponse speedVSAvoidsystem-wide energy optimization

Core Design Contradiction:

SpeedVSLoss of energy

Solution Approach 1:

The system implements feedback by having Resource Manager Proxies continuously monitor local computer execution states and report them to the centralized Resource Manager. The Resource Manager uses this feedback to make informed power configuration decisions that optimize system-wide energy consumption while respecting local response requirements. The feedback loop enables global optimization without sacrificing local response speed.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

Each Resource Manager Proxy is given the authority and capability to make immediate local power configuration decisions based on its local computer's execution state. This local quality empowerment allows fast response times at each node while the centralized Resource Manager ensures these local decisions contribute to overall system-wide energy optimization through coordinated guidance.

Inventive Principle:

Principle #3Local quality

3Loss of information

If Resource Manager collects state information from all computers, then global application state awareness is improved, but communication overhead increases

Engineering Contradiction:

Improveglobal application state awarenessVSAvoidcommunication overhead

Core Design Contradiction:

Loss of informationVSLoss of energy

Solution Approach 1:

The Resource Manager collects only the specific execution state information that is necessary for making power configuration decisions, rather than gathering all possible data from each computer. This partial action approach maintains global application state awareness sufficient for energy optimization while minimizing communication overhead and energy consumption.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS10168751B2Method of executing an application on a distributed computer system, a resource manager and a distributed computer system

Publication Date: 2019.01.01 FUJITSU LTD

AI summary

A Resource Manager is provided for managing a plurality of computers. Each of the computers is operable in one of a plurality of power configurations, including a high-power configuration and low-power configurations. The Manager exchanges messages with a Manager Proxy in each computer. Responsive to notifications from Manager Proxies of changes in execution state of an application, the Manager determines a power configuration applicable for the computer of that Manager Proxy, or for a set of computers executing the application. The Manager indicates the power configuration to the Manager Proxy which then decides whether to implement a change.