Host-Scheduled FDP Writes for Low-Garbage-Collection Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional memory sub-systems face increased write amplification and reduced performance due to inefficient garbage collection processes, which are not adequately addressed by existing communication protocols between host systems and memory sub-systems, leading to degraded system performance and reduced endurance.
Innovation Solution
Implementing flexible direct placement (FDP) to allow the host system to specify data placement on reclaim units, coordinating with the memory sub-system to minimize garbage collection through host folding operations, reducing write amplification by ensuring data with similar lifespans are grouped for simultaneous deletion, and using minimal protocol changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional garbage collection processes are used in memory sub-systems, then storage resources can be reclaimed, but write amplification increases and performance degrades
Solution Approach 1:
The host system performs preliminary actions by scheduling write operations to complete before the memory sub-system performs garbage collection. The host system tracks its own write operations and coordinates completion times to ensure data is written and confirmed before GC begins, reducing the need for data copying during GC and thereby reducing write amplification.
Solution Approach 2:
The host system uses feedback from the memory sub-system about garbage collection status and storage resource availability to dynamically adjust its write scheduling. By monitoring GC progress and storage usage metrics, the host system can optimize its write operation timing to minimize interference with GC processes and reduce write amplification effects.
2Reliability
If garbage collection operations are performed frequently, then storage resources are maintained, but system performance and endurance are reduced
Solution Approach 1:
The host system performs preliminary data writing and confirmation operations before the memory sub-system initiates garbage collection. By scheduling writes to complete in advance and receiving confirmation that data is securely stored, the host system ensures that when GC occurs, minimal or no data copying is needed, thereby reducing the frequency and impact of GC operations on system endurance.
Solution Approach 2:
The host system takes self-service actions by independently tracking and managing its own write operations and coordinating their completion with the memory sub-system's garbage collection process. This self-coordination allows the host to optimize GC timing and reduce the burden on the memory sub-system, thereby extending system endurance without compromising storage resource maintenance.
3Productivity
If data placement directives are implemented, then data with similar lifespans can be grouped, but protocol complexity increases
Solution Approach 1:
The host system implements a universal data placement mechanism that works with existing storage protocols while adding FDP functionality. The same write command structure used in conventional protocols is enhanced to optionally include placement directives, allowing the system to maintain compatibility with existing protocols while achieving improved GC efficiency through multi-functional command handling.
Solution Approach 2:
The protocol complexity is managed by changing parameters within existing command structures rather than creating entirely new protocols. The host system modifies existing write command parameters to include optional placement directives, and the memory sub-system adjusts its GC behavior based on these parameter changes, thereby achieving improved GC efficiency without proportionally increasing protocol complexity.
Data Source
AI summary
A host system having at least one processing device to execute instructions configured to implement a garbage collection manager, which can: compute, based on write commands sent from the host system to the memory sub-system according to a protocol of flexible direct placement (FDP) to store data of storage space tenants, an indicator of usage level of reclaim units in the memory sub-system; identify, in response to a determination that the indicator meets a condition, at least one reclaim unit in the memory sub-system having valid data; and communicate with the storage space tenants having the valid data in the at least one reclaim unit to cause the storage space tenants to perform operations that reduce garbage collection in the memory sub-system in reclaiming storage resources of the at least one reclaim unit.