Why does a connection failure occur during CLB stress testing?

Normally, when using the Linux operating system as the stress testing simulation client, there will not be a connection failure before the performance of the stress test reaches the CLB limit.

However, when using Windows as a stress testing simulation client, there may be TCP connection failure issues. This is because under the pressure test scenario, the Windows system will quickly reuse the client IP and port to establish TCP connections, but on the backend Linux service nodes of the tested CLB, the TCP connections established with the same source address and port may not have been fully released. If at this time the “sequence number of the new connection” is greater than the “sequence number of the existing connection”, the Linux service node will think that the SYN request of the new connection is a retransmission of the existing connection, which will cause the new TCP connection to fail to establish.

So when doing stress testing on the CLB, please try to use Linux as the stress testing simulation client. If you need to use the Windows system as the stress testing simulation client, you need to add an option for TCP timestamps in the system registry. This option is not activated by default in the Windows system. For specific configurations, please refer to the Windows help document.