Gotihub
Quant Edge Exchange
Powered by Apurba Labs (Gotihub)
Design Decisions

Great Systems are Built from Decisions

Quant Edge Exchange separates transient traffic from authoritative business truth, allowing each workload to be optimized independently.

Decision 01 // High-Throughput IngestionDynamoDB

Why DynamoDB for Traffic?

Problem

Millions of bid events can arrive continuously across regions. Most of these records are temporary and do not represent business truth.

Decision

Store transient auction traffic in Amazon DynamoDB.

✓ High write throughput
✓ Horizontal scalability
✓ Low operational overhead
✓ Fast regional aggregation
Decision 02 // Absolute ConsistencyAurora DSQL

Why Aurora DSQL for Settlement?

Problem

Winning bids become financial outcomes. Incorrect settlements cannot be tolerated.

Decision

Store authoritative settlement records in Aurora DSQL.

✓ Serializable transactions
✓ Strong consistency
✓ Financial auditability
✓ Multi-region durability
Decision 03 // Separation of ConcernsIsolation Core

Why Separate Traffic From Truth?

Problem

Treating every event as a transaction creates unnecessary coordination overhead.

Decision

Traffic and business truth are stored in specialized systems.

✓ Reduced contention
✓ Better performance
✓ Clear ownership
✓ Lower consistency cost
Decision 04 // Lock-Free Conflict Backoff PipelineOCC Jitter

Why OCC + Full Jitter?

Conflict Scenario

Worker A settle(slot_123)
Worker B settle(slot_123)
Serialization Conflict
Attempt #1: ✕ Attempt #2: ✕ Commit: ✓

Decision

Use Optimistic Concurrency Control with randomized exponential backoff to recover safely from write conflicts.

✓ Conflict recovery
✓ Fair retry distribution
✓ Reduced retry storms
✓ Stable throughput
Decision 05 // Strategic Workload ComparisonTopology Proof

Why Not Use One Database?

Single Database

✕ Transaction contention
✕ Higher coordination cost
✕ Expensive scaling
✕ Mixed workload patterns

Specialized Storage

✓ Traffic optimized
✓ Truth optimized
✓ Independent scaling
✓ Better cost profile
Key Architectural Insight

Not every event deserves global consistency.

Auction traffic is abundant. Financial truth is scarce. Treat them differently.

Traffic Events Log ──► Amazon DynamoDB
Financial Consensus Truth ──► Amazon Aurora DSQL

The cost of coordination should match the value of the data.