Background
A SaaS platform’s database architecture, designed for a single-region deployment, began experiencing intermittent data consistency errors as the company expanded to multi-region deployment to meet customer latency requirements.
The Challenge
The team needed a replication approach that maintained strong consistency for critical financial records while keeping read latency low across regions — a combination standard off-the-shelf replication configurations could not achieve for this workload.
Technological Uncertainty
It was not known in advance whether any replication strategy could meet both the consistency and latency requirements simultaneously for this specific transaction pattern, or whether a fundamental trade-off would make one requirement unachievable.
Experimental Development
The team systematically tested multiple consistency models and conflict-resolution strategies under simulated multi-region load, measuring latency, consistency violations, and failure recovery behaviour for each configuration.
What Failed?
An initial eventual-consistency approach met latency targets but produced consistency violations on financial records under concurrent writes, and a subsequent strict-consistency configuration eliminated violations but pushed latency above acceptable limits for two regions.
Technological Advancement
The team developed a hybrid consistency model that applied strict consistency selectively to financial-record operations while allowing eventual consistency elsewhere, generating new architectural knowledge specific to the platform’s transaction patterns.
Potentially Relevant SR&ED Activities
- Systematic testing of consistency models under simulated multi-region load
- Development of a selective, hybrid consistency architecture
- Failure-recovery and latency benchmarking across iterations
What Would Generally Not Qualify
Deploying the finalized architecture to additional regions using the same, already-validated configuration would not itself involve further eligible experimental development.
Documentation
Load-testing results across configurations, architecture decision records documenting the trade-offs considered, and incident postmortems from the failed approaches would support this claim.
