SQL Server performance & reliability
Billing Performance Engineering
I led SQL Server and query-plan diagnosis that cut average bill-generation time from 80 to 40 minutes in comparable staging runs and removed the deadlocks and process failures observed there.
01 / Context
What needed to change
A SQL Server-backed billing process was hitting contention and scaling unreliably under parallel execution.
02 / Role
What I owned
I was the primary engineering owner, working with the chief architect and CTO and documenting the diagnosis and changes for the wider engineering team.
03 / Constraints
Scope and limits
The comparison used the same staging environment, dataset, configuration and parallelism. The exact run count and workload size were not retained, so the result is a staging observation rather than a production benchmark.
04 / Approach
How I approached it
- Diagnosed database contention and query plans under parallel execution.
- Iterated on database and query-level changes for more than a month.
- Held the environment, data, configuration and settings constant for the before-and-after comparison.
- Documented the diagnosis and changes for engineering follow-through.
05 / Verification
How the result was checked
With the environment, dataset, configuration and parallelism held constant, average runtime fell from 80 to 40 minutes; the previously observed deadlocks and process failures did not recur.
- 80 → 40 minutes
- 50% observed runtime reduction
06 / Outcome
What changed
Average bill generation fell from 80 to 40 minutes in the comparable staging runs. The deadlocks and process failures seen before the changes did not recur in those runs.
07 / Lessons
What I carried forward
A performance result is only useful when its comparison conditions are clear and reliability is measured alongside speed.