Epoch Synchronization for Distributed RAID Read Latency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing RAID systems face performance limitations in flash memory due to the asymmetry in read, write, and erase times, and the need for coordinated operations to prevent interference with low-latency reading, especially in large-scale distributed memory systems where maintaining synchronization across multiple memory units is complex.

Innovation Solution

A memory storage system with synchronized memory modules using a synchronization signal generator, where each memory unit adjusts its clock oscillator to maintain epoch synchronization, allowing coordinated operations that prevent simultaneous erase or write operations within a RAID group, thereby achieving low-latency read performance through 'erase hiding' or 'write hiding' techniques.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If memory modules operate independently without synchronization, then each module can perform operations autonomously, but simultaneous erase or write operations interfere with low-latency read operations

Engineering Contradiction:
Improvedata protectionVSAvoidread latency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary scheduling of erase and write operations to identify time intervals when no such operations are pending on any memory module. Read operations are then scheduled during these pre-identified safe time intervals, ensuring low-latency reads without interference from erase or write operations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system continuously monitors the state of all memory modules to determine when erase and write operations are complete. This feedback information is used to dynamically update the schedule of safe read time intervals, allowing the system to adapt to changing operation states and maintain optimal read performance.

Inventive Principle:
Principle #23Feedback

2Loss of time

If erase and write operations are scheduled to prevent interference with reads, then read latency is reduced, but the coordination complexity across distributed memory units increases

Engineering Contradiction:
Improveread latencyVSAvoidsynchronization complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The system divides the distributed memory array into multiple independent RAID groups, where each group can be scheduled independently. This segmentation allows the coordination complexity to be contained within smaller subsets of memory modules rather than requiring system-wide coordination, reducing the overall synchronization complexity while maintaining low read latency.

Inventive Principle:
Principle #1Segmentation

3Adaptability or versatility

If memory modules are distributed across multiple independent units, then system scalability is improved, but maintaining epoch synchronization across all modules becomes more difficult

Engineering Contradiction:
Improvesystem scalabilityVSAvoidsynchronization coordination
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary determination of epoch intervals and schedules erase/write operations within these pre-established time boundaries. By pre-defining the temporal structure of operations across all distributed modules, the system enables scalable coordination without requiring complex real-time negotiation between modules.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses periodic epoch intervals to structure memory operations across all distributed modules. Each module operates according to this common periodic timeline, with erase, write, and read operations scheduled in regular intervals. This periodic structure provides a simple yet effective synchronization mechanism that scales to large distributed systems.

Inventive Principle:
Principle #19Periodic action

Data Source

PatentEP2870538B8Synchronization of a dispersed raid group
Publication Date: 2019.11.27 VIOLIN SYSTEMS LLC

AI summary

In a memory system where memory units may be separated from each other so as to operate substantially independently, the coordination of related memory operations between such units may be by synchronization of an epoch of time and the start of an epoch of time with a common synchronization source. The source may be distributed directly to each of the memory modules of a memory unit, or through an intermediate synchronization circuit of a memory unit that is common to the modules. Where the data is stored as a RAID stripe on a plurality of synchronized modules, the read and write or erase operations performed by the modules may be arranged such that the write operations or erase operations may not substantially affect the ability to promptly read the stored data of a RAID stripe.