Two NE40s were connected through E1-MPGroup link to carry NGN services. In traffic tests, the customer found that the traffic usage rate of E1 link was very poor. Packet loss occurred frequently when the traffic exceeded 30% of the total traffic of the link.
1. Careful analysis of the test process of the customer found that IPerf sent UDP traffic for tests.
2. Analysis of the principle for IPerf to send UDP traffic found the reason ― UDP burst traffic.
3. Analysis of the principle to implement NGN services found that NGN services actually used TCP traffic. The bandwidth used by actual services in the existing network can basically reach the interface bandwidth.
4. The customer was advised to use IPerf TCP traffic for tests and, basically, no packet loss occurred when traffic approximately equal to the interface bandwidth was sent.
5. Smartbit was used to the conduct the same test and, similarly, no packet loss occurred when traffic approximately equal to the interface bandwidth was sent.
The customer used UDP traffic for tests. Packet loss occurred frequently when the average rate did not exceed the interface rate.
Analyze the test software IPerf used by the customer. This software sends traffic in UDP and TCP modes. TCP is a connection-oriented service and the sending end can continue to send packets only after receiving the ACK message from the receiving end. Therefore, the traffic becomes steady after a period of time and the forwarding rate can basically reach the interface bandwidth. UDP is a connectionless service and the sending end keeps sending packets. For example, 1 M traffic is sent per second and the average traffic is 1 M/s. However, actual traffic is sent within over 10 ms, causing burst and the rate exceeds the maximum rate of the interface, leading to packet loss.
The customer used UDP traffic for tests, so the preceding phenomenon occurred. However, the existing network carries NGN services that are transmitted through TCP connections and using TCP traffic will not cause the preceding problem.