Banking backend / Ajay at Slice · previous employment
A distributed limit on a core banking path.
At Slice, Ajay engineered a distributed rate limiter for a core banking system.
Earlier experience1,000 TPS
Transactions per second at peak.
<90 ms
P99 latency for the limiter.
The system
A core banking request path.
The setting
The problem behind the work.
A core banking system needed to control request volume across more than one service instance while keeping the decision fast enough for a critical path.
The limiter had to enforce a distributed allowance under load without becoming a source of avoidable latency.
How it works
The workflow at a glance.
Simplified view of the workflow.
Contribution
What Ajay built.
Ajay's role
Ajay engineered the distributed rate limiter while working at Slice.
Approach
He designed the limiter for the banking path with both shared traffic control and response latency in view. Its implementation details remain within the former employer's system.
Design choices
What the work required.
Control traffic across instances
A distributed limit gives the service a consistent way to manage incoming load.
Keep the decision fast
The check sits on the request path, so its latency matters to the customer-facing operation.
Evaluate tail latency
P99 latency reveals slow requests that an average can hide.
Result
What changed.
The limiter handled up to 1,000 transactions per second at peak. Its reported p99 latency was under 90 ms.
Start a conversation