Sideband Interface for Stranded Data Recovery in Failed Nodes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In computing systems, data stored in non-volatile memory (NVM) of a failed node becomes inaccessible to other consumers, leading to 'stranded' data, even with emerging memory technologies that enable long-term storage after power removal.

Innovation Solution

A sideband interface connected to the DIMM controller allows data to be read and written through a management subsystem using standby power, enabling access to stranded data in NVM even when the node's main power is off, utilizing techniques like the Serial Presence Detect (SPD) interface and management buses for data migration and rebuilding.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Duration of action of stationary object

If non-volatile memory (NVM) is used for long-term data storage after power removal, then data persistence is improved, but data accessibility after node failure deteriorates

Engineering Contradiction:
Improvedata persistenceVSAvoiddata accessibility
Core Design Contradiction:
Duration of action of stationary objectVSEase of operation

Solution Approach 1:

A management device serves as an intermediary between the NVM DIMM and external systems. This management device includes a management processor and management interface that can be accessed through a sideband interface, allowing data to be read and written even when the main system is powered off or has failed. The management device enables data extraction from stranded NVM to external storage or migration to target nodes.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system separates the management functions from the main computing functions. The management device operates independently through a sideband interface, allowing data access operations to be performed separately from the main system operations. This segmentation enables the management processor to handle data extraction and migration tasks independently of the failed node's main processor.

Inventive Principle:
Principle #1Segmentation

2Ease of operation

If standard DIMM interfaces are used for memory access, then ease of data access is improved, but ability to access data after node failure deteriorates

Engineering Contradiction:
Improvedata access easeVSAvoiddata availability after failure
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The management device is designed to work with standard DIMM interfaces while adding specialized failure-recovery capabilities. The sideband interface provides a universal access method that works both during normal operation and after system failure, making the system multi-functional. The management processor can handle both routine data access and emergency data extraction through the same interface infrastructure.

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

3Loss of energy

If main power is removed from a failed node, then power consumption is reduced, but data extraction capability deteriorates

Engineering Contradiction:
Improvepower consumptionVSAvoiddata extraction capability
Core Design Contradiction:
Loss of energyVSEase of operation

Solution Approach 1:

The management device is designed to remain operational or be rapidly activatable after main power removal. The sideband interface and management processor are configured to maintain or restore functionality even when main power is cut, enabling data extraction to begin immediately without requiring full system power. This preliminary preparation of the management path allows energy-efficient operation while maintaining data access capability.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS10657016B2Recovering stranded data
Publication Date: 2020.05.19 HEWLETT PACKARD ENTERPRISE DEV LP
  • US10657016B2 patent drawing
  • US10657016B2 patent drawing
  • US10657016B2 patent drawing

AI summary

A node for a computing system may include a memory module, a main node processor and a management processor. The memory module may include a non-volatile memory, a module memory controller having a main bus interface for connection to a main bus and a management device providing access to the non-volatile memory through a sideband management bus. The main node processor is connected to the module memory controller and has a main bus interface for connection to a main bus. The management processor has a sideband interface for connection to the sideband management bus. The sideband management processor detects a failure of the node and, in response thereto, copies data from the non-volatile memory of the memory module to another node across the sideband management bus.