No relevant resource is found in the selected language.

This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy>

Reminder

To have a better experience, please upgrade your IE browser.

upgrade

Load Balancing Fails on an NE40 based on Its Three Next Hops, Causing the G9 Voice Quality Deterioration.

Publication Date:  2013-01-07 Views:  24 Downloads:  0
Issue Description

The G9 bearer network between two cities uses the single-layer networking. An NE40 is deployed in each city. The NE40s in different cities are directly connected through 21 E1 links. The E1 interfaces are bound to three MP-Groups with every seven E1 interfaces bound to an MP-Group. OSPF is running on the two NE40s to ensure that there are three equal-cost routes to the peer end. Per-packet load balancing is configured on the NE40s. The total bandwidths of the 21 E1 interfaces are abundant in carrying all G9 voice and signaling traffic according to the traffic planning. However, the voice quality is extremely poor during the peak hour at 20:00 every Friday night.

Handling Process

The services are load balanced through the three MP-Groups based on the 1.2:1:1 bandwidth ratio.

The on-site Huawei technical support personnel add another E1 member interface to the MP-Group with the bandwidth ratio 1.2. This scheme applies to a fixed service model because load balancing is performed using the hash algorithm with the destination address as the hash key. If the service model changes frequently, the MP-Group that takes over traffic with the 1.2 bandwidth ratio cannot be determined. Two or three MP-Groups use the same bandwidth. Incoming and outgoing calls are restricted based on 88% of the total bandwidths of the three MP-Groups on the UMG side.

Root Cause

The on-site Huawei technical support personnel check the core network side and find that on alarm is generated.

Then they check the packet loss on the NE40s during peak hours. Before they observe the statistics on the interfaces, they clear the statistics about all packets on the interfaces. They find that the number of lost packets on one MP-Group increases continuously while no packet loss occurs on the other two MP-Groups.

They check the transmission on the 21 E1 links and find that all E1 links are working properly. Therefore, the packet loss on one MP-Group is not caused by the E1 transmission failure.

They check the traffic statistics on the NMS and find that no matter how large the total traffic volume is on the network, the load ratio on the three MP-Groups remain 1.2:1:1.

Based on the further analysis, the NE40s cannot perform per-packet load balancing based on three next hops. The actual bandwidth ratio between the three MP-Groups is 1.2:1:1. Therefore, under the conditions of three MP-Groups with equal member interfaces, when the traffic rate exceeds 88.67% of total bandwidths of the three MP-Groups, one of the MP-Groups begins to discard packets.

Solution

Load balance Layer 3 traffic by specifying the sipxordip parameter.

Suggestions
Because NE40s perform load balancing using the hash algorithm with the destination IP address as the hash key, an NE40 can perform per-packet load balancing on two MP-Groups with the 1:1 bandwidth ratio. However, when there are three next hops, the NE40 fails to load balance the traffic based on the 1:1:1 bandwidth ratio. The actual bandwidth ratio is 1.2:1:1. Which next hop takes over the traffic forwarding depends on the destination IP address of the traffic.

END