Smart Memory Module P2P Live Migration Under Buffer-Full Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing live migration processes in data centers are complex and resource-intensive due to host software overhead, particularly in managing data consistency and performance during memory and storage transfers.
Innovation Solution
Implementing smart memory modules with peer-to-peer (P2P) communication capabilities, such as through Compute Express Link (CXL), to offload live migration operations directly between devices, minimizing host software involvement and optimizing data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If host software is involved in live migration processes, then data consistency and service continuity can be maintained, but system complexity and resource overhead increase significantly
Solution Approach 1:
The patent extracts the live migration operation from the host software domain and relocates it to the memory device domain. The memory device autonomously performs migration operations including data copying, buffer management, and flush operations, eliminating the need for complex host software intervention while maintaining data consistency through coordinated P2P communication.
Solution Approach 2:
The memory device is empowered with self-service capabilities to perform live migration operations independently. It manages its own buffer cache, initiates data transmission to target devices, and handles migration coordination without requiring host software to manage these complex operations, thereby reducing system complexity while ensuring reliability.
2Productivity
If host software performs data copying during live migration, then data can be transferred between systems, but processing speed is reduced due to software overhead
Solution Approach 1:
The patent replaces the mechanical software-based data copying process with a direct hardware-level P2P communication mechanism. Memory devices communicate directly with target memory devices through P2P channels, bypassing the software processing layer entirely, which dramatically increases data transfer speed and reduces migration time.
Solution Approach 2:
The patent introduces P2P communication channels as intermediaries between source and target memory devices. These dedicated communication paths enable direct data transmission without passing through host software, thereby eliminating software overhead and accelerating the migration process while maintaining data integrity.
3Reliability
If repetitive data copying is performed during live migration, then complete data transfer is achieved, but system resources are consumed excessively
Solution Approach 1:
The patent implements preliminary action by having the memory device maintain a buffer cache that pre-stores data before migration is needed. During live migration, data is copied from this pre-prepared buffer rather than being repeatedly read from and copied between storage media, significantly reducing system resource consumption while ensuring complete and accurate data transfer.
Solution Approach 2:
The patent uses partial action by implementing incremental migration where only the necessary portion of data is transferred at each step. The buffer cache mechanism allows the system to migrate data in optimized batches rather than performing exhaustive repetitive copying, reducing overall resource consumption while achieving complete data transfer.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A smart memory module (100) includes at least one memory device (110)connected to at least one channel (Dch), a controller (120) configured to control the at least one memory device (110), and a host interface circuit (130) configured to perform peer-to-peer, P2P, communication with an external device, and to receive, from a host device, a live migration request. The controller (120) is configured to perform, in response to the live migration request, a migration operation of transmitting buffer data to a target device through the P2P communication under a buffer-full condition.