Throughput is the rate at which completed work passes through a system. In software, architectural throughput depends on the components requests must pass through; improving a non-limiting component may not improve the system. Reliable delivery to users matters more than simply producing more work. section 11.4 section 1.1.1
Use the Theory of Constraints to improve throughput:
- Identify the component limiting the system, such as a shared database under peak load. section 11.4
- Exploit its existing capacity, then reduce unnecessary demands from other services. section 11.4
- If needed, elevate capacity; then identify the new constraint. section 11.4
A peak-load bottleneck
In a food-delivery platform, restaurant, order, and driver services share PostgreSQL. At peak load, their response times rise sharply while the Stripe-backed payment service changes little. The shared database is the likely constraint. Caching menus, moving driver updates to a time-series database, or batching notification queries can reduce its workload. section 11.4
Table 11.5 Checkout request latency breakdown by service
|
Service
|
Avg response (normal)
|
Avg response (peak)
|
Change
|
|
Restaurant Service
|
45ms
|
380ms
|
8.4x
|
|
Order Service
|
60ms
|
520ms
|
8.7x
|
|
Driver Service
|
30ms
|
290ms
|
9.7x
|
|
Payment Service
|
200ms
|
210ms
|
1.05x
|
Compare normal and peak response times across services to see which workloads degrade together. section 11.4































