Migrating Mission Critical Workloads from Mainframe to Open...
summary
U.S. Bank moves 60% of read traffic off the mainframe into Cassandra by piping mainframe CRUD events through an API/Kafka stream into Spark which atomically writes to Cassandra and serves reads from a Spring Boot domain layer. Latency falls from ~5 seconds to mostly 500 milliseconds and stays under the 1s SLA after tuning (partitions, continuous streaming, in‑DB checkpoints). The rollout includes journaling, hourly reconciliation, fail‑back to the mainframe, and runs on private‑cloud Kubernetes/VMs proving reliability, consistency and cost gains.
AI-written summary · about
chapters (20)
- 0:00Intro
- 0:46Background
- 1:55US Bank
- 4:04Digital Transactions
- 6:53Mainframe Architecture
- 13:46Mainframe Learnings
- 16:16Mainframe Pipeline
- 18:30Missing Events
- 21:03Summary
- 22:51Missing Events Detection
- 25:14Data Reconciliation
- 28:14Highlights
- 29:43How long was the migration effort
- 30:26How long did it take to build the streaming pipe
- 33:14What is a materialized view
- 37:54Kafka
- 39:54UI Layer
- 42:14Why its not tangible
- 43:04Why its declining
- 43:59Private Cloud
related talks
32mMigrating Workloads from Mainframes to Cassandra - Problems, Experience and Learnings
35mThe What and Why of Cassandra Datamodelling at Scale
46mEvolving Itaú-Unibanco's Architecture: From Mainframe to Java
27mMigrating from a Monolith to Microservices
26mBuilding Financial Systems on Eventually Consistent DBs
53mGo at Mainframe Scale
35mMission Critical Data, and Evaluating the Community in OSS Tech
24mMigrations under Production Load: How to Switch...