Pluggable Spare Memory Recovery for Failing DRAM Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud computing systems face reduced computing capacity due to DRAM module failures, which cause memory errors and necessitate nodes to be taken offline for repair, leading to resource inefficiencies and potential service disruptions.
Innovation Solution
Implementing a CXL Pluggable Spare Memory Module (CPSMM) that mirrors data from failing DRAM to a spare module, allowing continuous operation while the faulty module is replaced, utilizing CXL protocols for memory pooling and quiescing capabilities to manage traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a DRAM module fails, then memory errors occur and the computing node becomes unstable, but taking the node offline for repair reduces computing capacity and productivity
Solution Approach 1:
The system performs preliminary actions by detecting DRAM module failures through predictive failure analysis before the modules completely fail. This early detection enables proactive replacement scheduling that minimizes impact on computing capacity while maintaining system availability.
Solution Approach 2:
The patent introduces an intermediary replacement process where failing DRAM modules are swapped out and replaced with new modules while the computing node remains operational. This intermediary replacement mechanism allows the system to maintain computing capacity while improving reliability through continuous operation.
2Reliability
If virtual machines are migrated to another node for memory repair, then the failing node can be taken offline, but this requires excess computing resources and increases device complexity
Solution Approach 1:
The patent extracts the failing DRAM module from the computing node and replaces it with a new module, separating the memory replacement operation from the computing node's operational state. This extraction approach eliminates the need for virtual machine migration and excess computing resources while maintaining system availability.
3Reliability
If the computing node is placed in out for repair status, then memory errors are prevented, but this prevents workload execution and reduces productivity
Solution Approach 1:
The system performs preliminary failure detection and schedules replacement before the DRAM module completely fails, allowing the computing node to continue executing workloads without interruption. This preliminary action maintains both memory stability and productivity simultaneously.
Data Source
AI summary
Systems and techniques for live memory recovery using a pluggable memory module are described herein. It may be detected that a spare memory module has been inserted into a computing device based on a signal transmitted from the spare memory module. The spare memory module may be initialized. A dynamic random-access memory (DRAM) module of the computing device may be identified that is predicted to fail. Freeze instructions may be transmitted to a virtual machine manager to pause virtual machines executing on the computing device. Memory data may be transferred from the DRAM module to the spare memory module. Memory addresses may be remapped from the DRAM module to the spare memory module. Unfreeze instructions may be transmitted to the virtual machine manager to resume the virtual machines executing on the computing device.