Why does my round-robin (mode 0) bond lose half of the traffic destined for it?

博客围绕Red Hat Enterprise Linux 4、5、6系统中Mode 0绑定设备展开。使用该模式时，一半流量莫名丢失。根源是交换机未配置合适Etherchannel，无法适应绑定行为致MAC地址抖动、流量路由错误。决议是配置Etherchannel，诊断症状有应用缓慢、回复丢失等。

https://access.redhat.com/solutions/342453

SOLUTION 已验证 - 已更新 2013年五月20日23:54 -

English

环境

Red Hat Enterprise Linux 6
Red Hat Enterprise Linux 5
Red Hat Enterprise Linux 4
Mode 0 (round-robin) bonding device

问题

Using a mode 0 bond, when using a tool such as tcpdump or ping, half of all traffic appears to be lost inexplicably.

决议

From the bonding.txt file:

Raw

    The balance-rr, balance-xor and broadcast modes generally
require that the switch have the appropriate ports grouped together.
The nomenclature for such a group differs between switches, it may be
called an "etherchannel" (as in the Cisco example, above), a "trunk
group" or some other similar variation.

Ensure that the switch has an appropriate Etherchannel configured to support a mode 0 bond - if it does not, this must be configured for the bond to work properly, otherwise there is likely no return path to the server resulting in up to half of all traffic lost.

根源

If a switch is unaware of a mode 0 bond on a system, it will likely see a flapping MAC address as each of the interfaces on the server repeatedly ARP out their respective addresses with varying IP's. This can cause traffic to be lost on the return-trip, as traffic is incorrectly routed or dropped by the switch as it is not configured to accommodate the behavior of the bond, necessitating the need for an etherchannel on the switch which will allow this behavior.

诊断步骤

Symptoms associated with this occurring have been application slowness, lost ping replies when contacting the server, and loss of traffic seen in tcpdump from the target server to a foreign host.