How Multiplexing Boosts .NET Core to Achieve Million‑Level RPS

1. Problems of Single Path

Each request‑response pair occupies a dedicated connection, monopolizing network read/write resources. Idle time wastes bandwidth, and the I/O overhead limits the achievable requests per second (RPS), causing performance bottlenecks and higher resource consumption when scaling.

2. Advantages of Multiplexing

Multiplexing allows multiple request‑response streams over a single connection, dramatically reducing the number of connections, consolidating I/O operations, and improving overall network throughput and latency.

3. Achieving Million‑Level RPS with Multiplexing

The author implemented a multiplexing test on .NET Core, aiming for single‑service throughput of one million RPS with low latency. The test workflow includes integrating multiplexing into BeetleX, which automatically merges messages into a single buffer to reduce I/O reads and writes.

4. Test Message Structure

Protobuf is used as the serialization format for request and response messages.

4.1 Request Message

4.2 Response Message

4.3 Server Processing Code

4.4 Service Response Object

Received messages are queued for processing; the server records request‑response times and total message counts.

4.5 Client Request Code

4.6 Client Test Initiation Code

The test runs with ten concurrent connections, reusing them for request‑response multiplexing.

5. Test Configuration

Two Alibaba Cloud servers with 12‑core CPUs (6 physical cores, 12 threads) running Ubuntu 16.04 and .NET Core 2.14 were used.

6. Test Results

6.1 Client Statistics

6.2 Server Statistics

6.3 Bandwidth Statistics

With ten multiplexed connections, the system handled up to 1,000,000 responses per second, most responses latency between 1–3 ms.

6.4 Test Code

https://github.com/IKende/BeetleX/blob/master/samples/MultiplexingConnectionTest.zip