Short-Connection Server Pressure Testing Methodology and Transaction Rate Evaluation

With evolving business scenarios, server performance has become increasingly critical, prompting continuous optimization of backend implementations. While long‑connection pressure testing is well known, this article introduces a short‑connection pressure testing approach suitable for servers where each client request opens and closes a TCP connection immediately after the transaction.

In long‑connection testing, a single TCP connection handles multiple transactions, and the maximum concurrent users are measured by the number of simultaneous TCP connections (Concurrent TCP Connection Capacity). This method also applies to gateway devices such as firewalls.

Short‑connection testing, however, treats each client request as an independent transaction with the connection closed (HTTP header Connection: close ). To evaluate the maximum concurrent users for such servers, the article recommends using the Transaction Rate metric defined in RFC 3511 (Maximum HTTP Transaction Rate).

The business scenario illustrated shows clients repeatedly opening TCP connections (open), sending HTTP GET/POST requests (blue arrows), receiving responses (purple arrows), and then closing the connections (red arrows). The maximum concurrent user count is derived from the highest achievable Transaction Rate combined with the actual business workload.

For a target of 10 k concurrent users, the article outlines two phases:

Client onboarding phase: 5 initialization interfaces, each generating 1 transaction per second, leading to a Transaction Rate of 5 /s. The required Transaction Rate is calculated as 5 × 10 000 / expected onboarding time.

Steady‑state phase: 5 business interfaces each invoked every 10 seconds, yielding a required Transaction Rate of (5 × 10 000) / 10 = 5 000 /s.

Figures (included as images) demonstrate the calculation results for different onboarding time windows, showing that achieving simultaneous onboarding of all clients would demand at least 50 000 transactions per second.

The conclusion emphasizes that short‑connection servers should be evaluated using Transaction Rate rather than traditional concurrent connection metrics, providing a clear and accurate view of performance requirements.

Glossary:

Transaction – a single HTTP request/response pair.

Long connection (persistent connection) – one TCP connection handling multiple transactions.

HTTP Transaction Rate – the maximum number of transactions per second a device under test can sustain, as defined in RFC 3511.

Reference: http://www.faqs.org/rfcs/rfc3511.html