Hierarchical Event Correlation for Supercomputer Monitoring

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current monitoring systems for supercomputers face processing delays due to the increasing complexity and number of events generated, leading to potential risks from undetected major incidents, as existing correlation engines are not scalable and often result in ad-hoc adaptations.

Innovation Solution

A modular monitoring system with a hierarchical structure that includes correlation engines, sensors, and flexible protocol compliance, allowing for dynamic configuration and distribution of modules to manage events efficiently, with features like load balancing and filtering to prioritize critical events and prevent system overload.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a traditional correlation engine is used to monitor supercomputer events, then the system can process events, but processing delays occur and the system cannot scale to handle increasing event volumes

Engineering Contradiction:
Improveevent processing speedVSAvoidmonitoring system architecture
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The monitoring system is divided into multiple independent correlation modules, each handling a specific subset of events or a particular correlation function. These modules operate in parallel, distributing the processing load and eliminating the single-point bottleneck of traditional correlation engines. Each module can be independently scaled, maintained, and configured based on specific monitoring needs.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a temporal dimension to event processing by implementing event buffering and time-window-based correlation. Events are not processed immediately but are buffered and correlated within defined time windows, allowing the system to handle burst loads smoothly and perform retrospective analysis. This transforms the processing model from synchronous to asynchronous, significantly improving throughput.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If the monitoring system processes all events in real-time, then detection accuracy improves, but system overload occurs and processing delays increase

Engineering Contradiction:
Improveincident detection accuracyVSAvoidevent processing throughput
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system implements selective event processing where not all events are correlated with the same depth or immediacy. Critical events triggering potential incidents receive full real-time correlation processing, while routine events use simplified correlation rules or are processed in batches. This partial action approach ensures reliable incident detection while maintaining high processing throughput by avoiding unnecessary full-correlation overhead for every event.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent implements continuous event buffering and sliding time-window correlation that operates without interruption. The correlation process continues seamlessly across event boundaries and system state changes, ensuring no incident detection gaps. This continuous operation maintains high reliability while the buffered approach allows the system to absorb and process events at a sustainable rate, preventing overload.

Inventive Principle:
Principle #20Continuity of useful action

3Measurement precision

If the monitoring system is customized for specific supercomputer configurations, then monitoring accuracy improves, but adaptability to different configurations decreases

Engineering Contradiction:
Improveevent correlation accuracyVSAvoidconfiguration flexibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

Each correlation module is designed with configurable parameters and rules that can be locally adjusted to match specific supercomputer configurations, rack layouts, or event types. The system allows fine-grained customization at the module level while maintaining a standardized overall architecture. This enables accurate correlation for each specific deployment scenario without requiring system-wide redesign.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The correlation modules are designed as universal, configuration-driven components that can handle multiple event types and correlation scenarios through programmable rules. A single module instance can be configured to monitor different supercomputer architectures, cooling systems, or event patterns by loading appropriate correlation rules and parameters. This universal design provides both accuracy for specific configurations and adaptability to new configurations without custom development.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Productivity

If more correlation engines are added to handle increased event volume, then processing capacity increases, but system complexity and maintenance costs increase

Engineering Contradiction:
Improveevent processing capacityVSAvoidsystem architecture complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

Multiple correlation modules are merged into a unified distributed processing architecture that shares common resources such as event buffers, configuration databases, and alert management interfaces. The modules communicate through standardized protocols and can be dynamically added or removed without affecting the overall system structure. This merging approach increases processing capacity while avoiding the complexity multiplication that would result from completely independent correlation engine implementations.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3143505B1Architecture for correlating events for supercomputer surveillance
Publication Date: 2023.04.19 BULL SA
  • EP3143505B1 patent drawingFigure 1~2
  • EP3143505B1 patent drawingFigure 3

AI summary

The invention relates to a method for managing a supercomputer (SC) comprising a set of components (C1, C2, C3...C6) comprising a step of generating notifications (N) by sensors associated with said components; a step of processing said notifications by a hierarchy of modules, said notifications forming incoming events for the modules of lower hierarchical level of said hierarchy, and each module comprising at least one correlation engine for correlating incoming events and generating outgoing events transmitted to the modules of higher level within said hierarchy as incoming events; a step of generating alarms (A), on the basis of the outgoing events generated by the modules of higher hierarchical level within said hierarchy.