Metro Replication Path Selection for Single-Site Read I/O

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In equidistant uniform metro replication configurations, caching the same metadata pages on both sites results in inefficient cache utilization and performance overhead due to thrashing, especially when scaling systems for practical usage scenarios with many stretched and non-stretched volumes.

Innovation Solution

Implement a methodology to select a single optimal site for receiving and servicing read I/Os directed to a stretched volume, adjusting path states to ensure fairness in resource utilization and balancing node resources, thereby optimizing I/O processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If read I/Os are serviced by both sites in an equidistant uniform metro replication configuration, then cache redundancy is provided, but cache utilization efficiency deteriorates due to thrashing

Engineering Contradiction:
Improvecache redundancyVSAvoidcache utilization efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments the read I/O workload by directing read requests to only one site (the owner site) rather than allowing both sites to service reads. This is achieved through path state management where the host is directed to send read I/Os exclusively to the owner site, eliminating cache thrashing while maintaining cache redundancy for fault tolerance

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces asymmetry in the previously symmetric equidistant uniform configuration by designating one site as the owner site that services all read I/Os, while the other site becomes a non-owner site that does not service reads. This asymmetric role assignment resolves the cache thrashing problem while maintaining the benefits of synchronous replication

Inventive Principle:
Principle #4Asymmetry

2Reliability

If both sites service read I/Os, then availability is improved, but system performance deteriorates due to resource imbalance and thrashing

Engineering Contradiction:
Improvesystem availabilityVSAvoidsystem performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments read I/O servicing responsibilities by assigning the owner site as the sole handler for read requests. This segmentation eliminates the performance degradation caused by both sites competing to service reads, while write I/Os continue to be replicated to both sites maintaining availability

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by allowing each site to have different roles (owner vs. non-owner) with respect to read I/O servicing. The owner site has the quality of servicing reads while the non-owner site does not, optimizing resource utilization at each location based on its role in the replication configuration

Inventive Principle:
Principle #3Local quality

3Device complexity

If path states are not managed, then configuration simplicity is maintained, but resource utilization fairness deteriorates

Engineering Contradiction:
Improveconfiguration simplicityVSAvoidresource utilization fairness
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-establishing path states (preferred, non-preferred, or equal) before I/O operations begin. This preliminary configuration enables the system to direct read I/Os to the owner site and write I/Os to both sites, ensuring fair resource utilization without requiring complex runtime negotiations or dynamic path selection

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS12498985B2Techniques for path selection in an equidistant uniform metro replication configuration
Publication Date: 2025.12.16 DELL PROD LP
  • US12498985B2 patent drawing
  • US12498985B2 patent drawing
  • US12498985B2 patent drawing

AI summary

In at least one embodiment, a method can include: establishing a bi-directional synchronous replication configuration for a stretched resource configured across first and second sites; determining a first difference between a first maximum normalized total I/O workload of the first site and a first current normalized total I/O workload of the first site without including read I/O workload of the stretched resource; determining a second difference between a second maximum normalized total I/O workload of the second site and a second current normalized total I/O workload of the second site without including read I/O workload of the stretched resource; and selecting, in accordance with the first difference and the second difference, only one of the first site and the second site to receive and service subsequent read and write I/Os directed to the stretched resource. The configuration can be equidistant with respect to a host and the stretched resource.