How to Define “Excellent” QPS Benchmarks for System Capacity Planning

Industry Benchmarks (2024 Practice)

Reference benchmarks for QPS excellence are illustrated below.

Key Evaluation Dimensions

1. Business Type Determines Baseline

Reading‑type applications: QPS > 5,000 is considered excellent.

Payment systems: QPS > 15,000 and success rate ≥ 99.99%.

Real‑time bidding systems: QPS > 100,000 with average latency < 10 ms.

2. Response‑Time Requirements

3. Resource‑Efficiency Standard

An excellent system must satisfy all three conditions simultaneously:

if (QPS > 10,000) and (single‑request CPU time < 5 ms) and (single‑request memory < 50 KB):
    qualifies as excellent system

Performance Golden Triangle Model

The model balances three dimensions: throughput, latency, and resource consumption.

Specific Metric Requirements

1. Load‑Test Performance

Run a 24‑hour load test at target QPS; error rate must be < 0.01%.

CPU utilization should stay between 60 % and 70 % (reserve ~30 % headroom).

Memory usage fluctuation should be within ±5 %.

2. Stability Indicators

3. Cost Efficiency

Per‑request cost ≤ $0.0001 in a cloud environment.

A cluster that supports 100 k QPS should not exceed 50 nodes of standard 8‑core / 16 GB configuration.

Architectural‑Level Recommendations

Thresholds for layered architecture capabilities are shown below.

Breakthrough Optimization Strategies

When QPS reaches a bottleneck, consider the tactics illustrated.

Real‑World Cases

Ant Financial: self‑built distributed architecture achieved 610,000 QPS during Double‑11 2023.

Twitter: adopted the Finagle framework to support >800,000 QPS for tweet distribution.

Netflix: Zuul gateway routes 1 million QPS video requests.

Final Conclusions

An excellent system should handle at least 120 % of expected peak traffic while keeping key indicators within limits:

Error rate ≤ 0.1 %.

TP99 latency ≤ the business‑tolerated threshold.

Overall resource utilization ≤ 70 %.

Excellent Threshold Calculation Formula

Excellent QPS = (Industry benchmark) × (1 + Business complexity coefficient)

Complexity Coefficients

Pure in‑memory computation: 0.8

Three‑level service calls: 1.2

Transactional consistency: 1.5

Assessments should be based on actual stress‑test results and continuously maintain the target QPS without degradation.

Source: https://blog.csdn.net/zybsjn/article/details/148719905