Selective Memory Dumping via Reference Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In modern data processing systems, capturing diagnostic information from memory dumps becomes increasingly costly and resource-intensive as memory sizes grow, with larger dumps requiring significant memory, compute resources, and network bandwidth, and often necessitating system quiescing to ensure a consistent memory state.

Innovation Solution

The implementation of selective diagnostic dumping techniques that track memory use by determining the unreferenced-interval count for sets of contiguous frames, allowing for the exclusion of infrequently accessed memory sets from diagnostic dumps based on a threshold, thereby reducing the size and resource requirements of memory dumps.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If full memory dumps are captured for diagnostic purposes, then complete diagnostic information is obtained, but memory requirements and processing overhead increase significantly

Engineering Contradiction:
Improvediagnostic information completenessVSAvoidmemory requirements
Core Design Contradiction:
Loss of informationVSQuantity of substance

Solution Approach 1:

The patent extracts only the necessary diagnostic information from memory by identifying and dumping only those memory pages that were actually accessed during the crash condition. The system uses reference tracking to determine which pages contain relevant diagnostic data and excludes unused pages, thereby reducing the overall memory footprint of the dump while preserving essential diagnostic information.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies local quality by differentiating between memory pages based on their access characteristics. Instead of treating all memory uniformly, the system identifies specific pages with high reference counts (indicating active use) and prioritizes their inclusion in the dump, while excluding pages with low or zero references. This selective approach optimizes the diagnostic dump for relevance while reducing size.

Inventive Principle:
Principle #3Local quality

2Loss of information

If larger memory dumps are generated, then more diagnostic data is available, but compute resources and network bandwidth requirements increase

Engineering Contradiction:
Improvediagnostic data availabilityVSAvoidcompute resources
Core Design Contradiction:
Loss of informationVSPower

Solution Approach 1:

The patent performs preliminary tracking of memory references during normal system operation, maintaining reference counts for each memory page before a crash occurs. This advance preparation enables the system to quickly identify which pages need to be included in a diagnostic dump when a crash is detected, avoiding the need to scan entire memory spaces during the crash handling process and significantly reducing compute resource requirements.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If system quiescing is implemented to ensure consistent memory state, then diagnostic accuracy is improved, but system stability and operational continuity are compromised

Engineering Contradiction:
Improvememory state consistencyVSAvoidsystem stability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent enables the memory subsystem to self-identify and self-select the pages that need to be dumped based on pre-tracked reference counts. This self-service mechanism eliminates the need for external system quiescing or complex coordination between different system components, as the memory manager itself determines which data is relevant and prepares the dump independently, maintaining system stability while ensuring diagnostic accuracy.

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS20240211334A1Selective diagnostic dumping of memory contents based on tracking memory use
Publication Date: 2024.06.27 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • US20240211334A1 patent drawing
  • US20240211334A1 patent drawing
  • US20240211334A1 patent drawing

AI summary

Techniques for selective memory dumping based on tracking memory use. For each set of contiguous frames of memory, whether a reference bit is set is determined. The reference bit is included in a control field of a frame of the respective set. Each set has an associated mapping from corresponding pages of a virtualized address space. An unreferenced-interval count is maintained for the respective set. The unreferenced-interval count represents an age since any frame of the respective set was last accessed. The unreferenced-interval count and the reference bit are cleared upon determining that the reference bit is set. If the unreferenced-interval count of a set does not exceed a threshold count, the set is included in a diagnostic dump of the memory when a condition for generating the diagnostic dump is met.