QPS vs TPS vs Concurrency: Uncovering the Real Difference in System Throughput

1. QPS

QPS Queries Per Second is the number of queries a single server can handle per second, measuring the maximum throughput of that server.

2. TPS

TPS Transactions Per Second counts the number of complete request‑response transactions a server processes each second.

3. Difference between QPS and TPS

TPS includes three steps: the user request, the server’s internal processing, and the server’s response. The number of these complete cycles per second equals TPS.

QPS is similar but counts every individual request to the server; a single page load may generate multiple QPS while representing one TPS.

Example: Visiting a page triggers three server requests, producing one TPS and three QPS. Analogy: A big eater can finish ten buns in a second (high TPS), while a slower eater can only finish one bun in the same time (low QPS), even though both are performing the same overall task.

4. Concurrency

Concurrency (or concurrency degree) is the number of requests a system can handle simultaneously, reflecting its load capacity. It can be derived from analyzing the number of log entries within one second.

5. Throughput

Throughput is the amount of work a system processes in a given time, commonly measured by QPS or TPS.

Key factors influencing throughput include CPU consumption per request, external interfaces, and I/O latency. Higher CPU cost, slower I/O, or inefficient interfaces reduce throughput.

Important parameters: QPS/TPS, concurrency, and average response time.

QPS/TPS: number of requests or transactions per second.

Concurrency: number of simultaneous requests/transactions.

Response time: usually the average response time.

Relationship: QPS (or TPS) = Concurrency / Average Response Time.

A system’s throughput is limited by the lower of its maximum QPS/TPS or concurrency; once either reaches its ceiling, additional load will degrade performance.

6. PV (Page View)

PV counts each page load or refresh as a separate view, typically obtained from daily access logs.

7. UV (Unique Visitor)

UV counts distinct users visiting a site within a day, derived by deduplicating logs based on unique identifiers.

8. DAU (Daily Active User)

DAU measures the number of users who actively use a product in a single day, similar to UV but focused on activity.

9. MAU (Monthly Active User)

MAU counts distinct active users over a month.

10. System Throughput Evaluation

When designing a system, consider CPU, I/O, and external service latency to estimate performance.

Beyond QPS and concurrency, daily PV is another dimension for capacity planning.

Typical approach:

Identify the system’s peak TPS and daily PV, which usually have a stable relationship (excluding holidays or seasonal effects).

Use stress testing or empirical estimation to determine the maximum TPS, then calculate the maximum daily throughput based on the TPS‑PV relationship.

11. Basic Concepts and Formulas for Software Performance Testing

From the user’s perspective, the critical metric is response time—the interval from an action (click, request) to the result being displayed.

From an administrator’s perspective, important performance aspects include response time, resource utilization (CPU, memory, database), scalability, maximum concurrent users, potential bottlenecks, hardware upgrades, and 24/7 availability.

From a developer’s perspective, considerations cover architectural design, database schema, code efficiency, memory usage patterns, thread synchronization, and resource contention.