Case Studies Banking
Top-4 American bank
Multi-region data hub at a 5-minute SLA
Global data hub remediation—incremental cross-region copy from Oracle (including Flexcube) held to a 5-minute SLA, with hub-and-spoke aggregation for residency and PII.
Apache Kafka · Kafka Connect · Oracle CDC · Apache Flink · Confluent
-
Customer
A top-4 American bank running a global data hub across regions and lines of business, under regulatory pressure on lag and uncontrolled copies of sensitive data.
Sources sat heavily on Oracle, including a legacy Flexcube footprint. The hub was meant to publish change once so consumers could trust freshness without every region mirroring everything.
-
Challenge
Copy data across regions incrementally at a bounded five-minute latency from Oracle change streams.
Oracle throughput, high cross-region RTT, and LogMiner limits made conventional low-latency replication hard. Kafka Connect and Oracle CDC were already in the path.
-
Work
Started with producer and consumer tuning. That bought headroom but did not fix the topology.
Moved to a low-latency hub-to-spoke design: regional hubs aggregate, then forward. Aggregation at the hub improved residency and PII posture, cut redundant multi-region copy, and held the five-minute SLA as a designed envelope.
-
Impact
- 5-minute incremental cross-region copy SLA—held under Oracle CDC / Connect constraints
- Hub-and-spoke topology after producer/consumer tuning alone proved insufficient
- Stronger residency and PII posture; redundant regional copies off the critical path
