Node Fault Recovery via Selective Hardware Reset

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Massively parallel supercomputers like the Blue Gene system face inefficiencies due to increased soft and hard failures, leading to system downtime and resource wastage, as failures in one node can render entire partitions unusable and require job abortion and restart.

Innovation Solution

A method and apparatus for fault recovery in a parallel computer system that uses a heartbeat monitor on a service node to detect node failures, allowing for selective resetting of non-network hardware portions of a failed compute node without interrupting executing jobs, enabling continued operation without restarting.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a node failure occurs in a massively parallel computer system, then the system must ensure high reliability, but the entire partition becomes unusable and requires job abortion

Engineering Contradiction:
Improvesystem reliabilityVSAvoidjob execution continuity
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments the compute node into distinct functional components: network hardware portion and remaining portion (processing unit, memory, etc.). The reset hardware can selectively reset only the network hardware portion while leaving the remaining portion operational, allowing the node to continue participating in job execution despite network-related faults. This segmentation enables partial recovery without complete node failure.

Inventive Principle:
Principle #1Segmentation

2Stability of the object's composition

If a node failure occurs, then system stability is compromised, but resetting the entire node causes job interruption and resource wastage

Engineering Contradiction:
Improvesystem stabilityVSAvoidsystem downtime
Core Design Contradiction:
Stability of the object's compositionVSLoss of time

Solution Approach 1:

The reset hardware is designed to reset only specific portions of the compute node (network hardware portion) rather than the entire node. This selective reset capability stabilizes the system by containing failures to specific components while maintaining overall node functionality and job execution, thereby reducing system downtime and avoiding resource wastage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of performing a complete node reset (excessive action), the system performs a partial reset of only the network hardware portion. This partial action is sufficient to recover from network-related faults while avoiding the unnecessary interruption of job execution and minimization of system downtime that would result from a full reset.

Inventive Principle:
Principle #16Partial or excessive action

3Reliability

If the network hardware portion is reset, then network connectivity is restored, but the entire node reset would interrupt executing jobs

Engineering Contradiction:
Improvenetwork connectivityVSAvoidjob execution continuity
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The compute node is segmented into network hardware portion and remaining portion, with separate reset control. The reset hardware can independently reset the network hardware portion to restore network connectivity while leaving the remaining portion (including executing jobs) undisturbed. This segmentation enables targeted recovery without collateral disruption to job execution.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2115588B1Fault recovery on a massively parallel computer system enabling node failure handling without ending an executing job
Publication Date: 2015.06.10 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • EP2115588B1 patent drawingFigure 1
  • EP2115588B1 patent drawingFigure 2
  • EP2115588B1 patent drawingFigure 3~4

AI summary

A method and apparatus for fault recovery of on a parallel computer system from a soft failure without endingan executing job on a partition of nodes. In preferred embodiments a failed hardware recovery mechanism on a service node uses a heartbeat monitor to determine when a node failure occurs. Where possible, the failed node is reset and re-loaded with software without ending the software job being executed by the partition containing the failed node.