IO Prediction Splitter for Asynchronous Data Replication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data replication systems face challenges in keeping pace with high data transaction rates at production sites, leading to potential data inconsistencies and system slowdowns, as they require acknowledgments for each transaction, which can result in latency and increased downtime during backups.
Innovation Solution
Implementing a prediction algorithm that identifies predicted IO locations, allowing the splitter to send IO down the stack without waiting for acknowledgments from the replication appliance, thereby reducing latency and enabling asynchronous processing, especially in sequential write scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If synchronous acknowledgments are required for each data transaction at the backup site, then data consistency is maintained, but the production site is forced to slow down due to latency
Solution Approach 1:
The patent applies preliminary action by predicting future I/O locations in advance using a prediction algorithm. The system pre-calculates where data will be written next based on sequential access patterns, allowing the backup site to prepare ahead of time. This eliminates the need for synchronous acknowledgments during actual data transactions, as the prediction has already been made and stored, thus maintaining data consistency while enabling high transaction rates at the production site.
2Reliability
If the backup site processes all data transactions synchronously, then complete logging is achieved, but the overhead causes the system to choke and force production site shutdown
Solution Approach 1:
The system performs preliminary action by using a prediction algorithm to determine future I/O locations before actual data transactions occur. This prediction information is stored and used to process transactions asynchronously, reducing the overhead of synchronous logging while maintaining completeness. The prediction mechanism allows the backup site to prepare logging operations in advance without blocking production site operations.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of a prediction algorithm and prediction information storage structure. This intermediary layer sits between the production site and backup site, translating sequential access patterns into predicted future locations. The intermediary enables asynchronous processing by providing advance information about where data will be written, thus reducing system overhead while maintaining logging completeness.
3Productivity
If asynchronous processing is implemented without prediction, then throughput is improved, but data inconsistencies occur due to backlog of un-logged transactions
Solution Approach 1:
The patent applies preliminary action by implementing a prediction algorithm that calculates future I/O locations in advance. This prediction information is stored and used to guide asynchronous processing at the backup site. By knowing where data will be written next, the backup site can process transactions asynchronously without creating backlogs, thus maintaining both high throughput and data consistency. The prediction mechanism ensures that asynchronous processing does not sacrifice reliability.
Data Source
AI summary
A computer implemented method, system, and computer program product for use in replication comprising receiving an IO at a splitter at a production site from a host, determining if the IO is to a predicted location, wherein a predicted location is a location predicted to correspond to a future write to an image on the production site, and splitting the IO based on the decision.