Read Latency Set Segmentation for SSD Tail Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern computer storage devices face increased read access latency due to write operations blocking read requests, leading to high tail latency and compromised Quality of Service (QOS), especially in non-volatile flash memory and SSDs, where existing methods fail to efficiently prioritize and duplicate data effectively.
Innovation Solution
A system that assigns storage segments to Read Latency Sets (RLS) with alternating read/write and read-only modes, prioritizes requests, and duplicates high-priority data across multiple RLSs for simultaneous writing and reading, utilizing RAM for immediate high-priority data access and metadata management to optimize storage and retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If write operations are performed in storage devices, then data is written to storage segments, but read access requests to the same memory space are blocked and read latency increases
Solution Approach 1:
The storage device divides storage space into multiple storage segments and assigns them to different Read Latency Sets (RLS). By segmenting the storage space and using multiple RLS with alternating work modes, the system allows read operations to proceed on read-only RLS while write operations occur on read/write RLS, thereby reducing read access latency without compromising write throughput.
Solution Approach 2:
The system dynamically switches work modes of Read Latency Sets between read-only and read/write modes. This dynamic mode switching allows the storage device to adapt to different operation requirements, enabling reads from protected read-only RLS while allowing writes to read/write RLS, thus resolving the conflict between read latency and write throughput.
2Loss of time
If data is duplicated across multiple storage segments to reduce read latency, then read access speed improves, but storage space consumption increases
Solution Approach 1:
The system applies data duplication selectively based on priority levels. High-priority data is duplicated across multiple Read Latency Sets to ensure low read latency, while low-priority data is stored in a single read/write RLS. This local quality approach ensures that duplication resources are allocated only where they provide the most benefit, optimizing both read latency and storage space utilization.
Solution Approach 2:
The system changes the duplication parameter based on data priority. High-priority data objects are duplicated across multiple RLS while low-priority data uses single-instance storage. This parameter change allows the system to optimize read latency for critical data while minimizing storage space consumption for non-critical data.
3Loss of time
If read operations are serviced during write operations, then read latency is reduced, but write operation reliability decreases
Solution Approach 1:
The system extracts read operations from the write operation path by directing reads to read-only RLS that are separate from read/write RLS. This extraction allows read operations to proceed independently without interfering with write operations, maintaining both low read latency and write operation reliability.
Solution Approach 2:
The read-only Read Latency Sets serve as intermediaries that buffer read access requests. By introducing this intermediary layer, the system allows reads to access stable data copies without blocking or being blocked by write operations on read/write RLS, thus maintaining both read responsiveness and write reliability.
Data Source
AI summary
A system and a method for reducing read latency of a storage media associated with at least one host computer, by at least one processor, may include assigning each storage segment of the non-volatile storage to a first Read-Latency Set (RLS) and a second RLS, wherein the first RLS is attributed a read/write mode and the second RLS is attributed a read-only mode; receiving read-requests and write-requests from the at least one host computer, wherein each of said requests is attributed a priority; and serving the received requests according to RLS work modes and according to the priority of each request.