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 Because the Mask Lengths Are Inconsistent.

Publication Date:  2013-01-07 Views:  74 Downloads:  0
Issue Description

As shown in the following topology (Only the key part of the topology is provided here), two NE40Es at the BSW site and two NE40Es at the SV site are connected through a dot1q sub-interface. There are transmission devices on the links between the BSW network and the SV network. The transmission device allocates bandwidths for the VLANs. 2 Gbit/s is allocated to voice traffic using L3VPN. The VPN instance name is 2G_Traffic. The VLAN IDs are 651, 652, 751, and 752. The allocated bandwidth is shared by the VLANs. For example, if VLAN 651 is allocated 110 Mbit/s, VLAN 651 used between SV-01 and BSW-01 shares the 110 Mbit/s bandwidth with VLAN 651 between SV-02 and BSW-02. To calculate and compare the bandwidth usage on the two sides, the traffic volume on the sub-interfaces on both sides must be summed.

OSPF is running on the NE40Es. Normally, there are four equal-cost routes destined for the peer routers. Traffic is load balanced on the four VLANs.

 

The preceding MRTG traffic diagram shows that the traffic distributions on the four VLANs differ to a large extent unexpectedly. The traffic volume on VLAN 651 and on VLAN 751 is similar while the traffic volume on VLAN 652 and on VLAN 752 is similar. But the traffic volume on VLAN 651 is nearly twice the traffic volume on VLAN 652.

The on-site Huawei technical support personnel collect traffic statistics on the interfaces, which matches the symptom. The result is shown as follows.

 

Handling Process

The on-site Huawei technical support personnel check the type of LPUs installed on the NE40Es at the SV site. The type is LPUF-10. Then they check the type of the two LPUs through which the NE40Es at the SV site are connected. One is an LPUF-10 and the other is an LPUF-21. Although the LPUF-21 may affect load balancing, the effect cannot be so obvious.

They check the routing table and find an abnormality in the routing table. There are not four but six routes in the routing table. The routes to the sub-interfaces on VLAN 651 and VLAN 751 are displayed twice.

The routing table on BSW-1 is shown as follows. Based on the further analysis, the routes originating from VLAN 651 have two different next hops. One is destined for the IP address of VLAN 651 on SV-1; the other is destined for the IP address of VLAN 651 on SV-2.

<BSW-BSC-NE40E-01>dis ip routing-table vpn-instance 2G_Traffic 10.67.0.0

Route Flags: R - relay, D - download to fib

------------------------------------------------------------------------------

Routing Table : 2G_Traffic

Summary Count : 6

Destination/Mask    Proto  Pre  Cost       Flags NextHop         Interface

  10.67.0.0/16  O_ASE  150  1            D   10.166.1.9      GigabitEthernet1/0/0.652

                    O_ASE  150  1            D   10.166.2.129    GigabitEthernet5/0/0.751

                    O_ASE  150  1            D   10.166.2.5      GigabitEthernet1/0/0.651

                    O_ASE  150  1            D   10.166.1.13     GigabitEthernet5/0/0.752

                    O_ASE  150  1            D   10.166.2.1      GigabitEthernet1/0/0.651

                    O_ASE  150  1            D   10.166.2.133    GigabitEthernet5/0/0.751

They check the configurations on the interfaces on both sites and find that the inconsistency in blocking.

 

They analyze the IP addresses and blocking. On VLAN 651 and VLAN 751, the four NE40Es are on the same network segment. However, on VLAN 652 and VLAN 752, the four NE40Es set up P2P connections.

Therefore, on BSW-1, VLAN 651 and VLAN 751 are connected to SV-1 and SV-2 respectively while VLAN 652 and VLAN 752 are both connected to SV-1. The connections are similar on other NE40Es. As a result, there are six equal-cost routes in the routing table. Two routes traverse VLAN 651 and two routes traverse VLAN 751. One route traverses VLAN 652 and one route traverses VLAN 752. When the NE40Es implement load balancing, traffic is evenly distributed over the six equal-cost routes. Because the total number of routes (4) traversing VLAN 651 and VLAN 751 is twice the number of routes (2) traversing VLAN 652 and VLAN 752, the traffic volume is distributed on the VLANs based on the route ratio.

Root Cause

When load balance ,The route mask must be the same.

Solution

Change the mask lengths to 30 bits.

Suggestions
Ensure that the mask lengths are consistent in load balancing. Be careful and note the details when you plan the network and deploy the configurations.
The possible causes for the problem are listed as follows:
The LPUF-20 or LPUF-21 cannot load balance traffic as expected.
The configurations on the NE40Es are incorrect or the protocols are not configured correctly
The heavy traffic is forwarded in per-session load balancing mode, transmitted intensely on some routes.

 

END