Time to practice. You will define a complete set of performance SLAs for the Banking Portal. This exercise forces you to think about metrics from a business perspective -- not just "what numbers look good" but "what numbers the business needs to succeed."
Open the Banking Portal and identify the 6 most critical operations: login, view balance, transfer funds, view transaction history, generate statement PDF, and account settings.
For each operation, define the p95 response time target. Remember: financial operations need stricter SLAs than browsing operations.
For each operation, define the maximum acceptable error rate. Financial transactions should be near-zero errors.
Define the expected concurrent user count for the bank. How many customers might be logged in simultaneously during peak hours?
Calculate the minimum throughput (RPS) for each operation using Little's Law. How many transactions per second does each endpoint need to handle?
Define system resource thresholds: max CPU, memory, database connections at peak load.
Define recovery SLA: after a traffic spike, how quickly should the system return to normal?
Write it all up in a table format like the SLA document example in this chapter.
| Operation | Think About | Suggested p95 |
|---|---|---|
| Login | First impression, user is impatient | < 1 second |
| View Balance | Users check frequently, must be instant | < 500ms |
| Transfer Funds | Critical financial operation, users wait | < 3 seconds |
| Transaction History | Could return 1000s of rows, needs pagination | < 2 seconds |
| Generate Statement PDF | Heavy operation, users expect delay | < 5 seconds |
| Account Settings | Infrequent, not time-critical | < 2 seconds |
Banking applications have compliance requirements. Transfer operations must have near-zero error rates (< 0.01%) because a failed transfer could mean lost money. A retry that accidentally processes twice could mean a double charge. Think about idempotency in your SLAs.
Key Point: Banking SLAs are stricter than e-commerce -- financial operations need near-zero error rates and compliance-aware thresholds