RHEL5下双网卡bonding做法

参考：
/usr/share/doc/kernel-doc-2.6.18/Documentation/networking/bonding.txt

将eth1和eth2做bonding为bond0

1:修改/etc/modprobe.conf，添加如下：
[root@ha01 network-scripts]# tail -2 /etc/modprobe.conf
alias bond0 bonding
options bond0 miimon=100 mode=1

关于mode见本文最下面。

2:在/etc/sysconfig/network-scripts/下创建ifcfg-bond0，ifcfg-eth1,ifcfg-eth2
[root@ha01 network-scripts]# cat ifcfg-bond0
DEVICE=bond0
BOOTPROTO=none
ONBOOT=yes
NETWORK=192.168.137.0
NETMASK=255.255.255.0
IPADDR=192.168.137.9
USERCTL=no
GATEWAY=
TYPE=Ernet
[root@ha01 network-scripts]# cat ifcfg-eth1
DEVICE=eth1
BOOTPROTO=none
ONBOOT=yes
MASTER=bond0
SLAVE=yes
USERCTL=yes
[root@ha01 network-scripts]# cat ifcfg-eth2
DEVICE=eth2
BOOTPROTO=none
ONBOOT=yes
MASTER=bond0
SLAVE=yes
USERCTL=yes

关于mode的部分

mode

         Specifies one of the bonding policies. The default is
         balance-rr (round robin).   Possible values are:

         balance-rr or 0

                 Round-robin policy: Transmit packets in sequential
                 order from the first available slave through the
                 last.   This mode provides load balancing and fault
                 tolerance.
         active-backup or 1

                 Active-backup policy: Only one slave in the bond is
                 active.   A different slave becomes active if, and only
                 if, the active slave fails.   The bond's MAC address is
                 externally visible on only one port (network adapter)
                 to avoid confusing the switch.

                 In bonding version 2.6.2 or later, when a failover
                 occurs in active-backup mode, bonding will issue one
                 or more gratuitous ARPs on the newly active slave.
                 One gratuitous ARP is issued for the bonding master
                 interface and each VLAN interfaces configured above
                 it, provided that the interface has at least one IP
                 address configured.   Gratuitous ARPs issued for VLAN
                 interfaces are tagged with the appropriate VLAN id.

                 This mode provides fault tolerance.   The primary
                 option, documented below, affects the behavior of this
                 mode.

         balance-xor or 2

                 XOR policy: Transmit based on the selected transmit
                 hash policy.   The default policy is a simple [(source
                 MAC address XOR'd with destination MAC address) modulo
                 slave count].   Alternate transmit policies may be
                 selected via the xmit_hash_policy option, described
                 below.

                 This mode provides load balancing and fault tolerance.

         broadcast or 3

                 Broadcast policy: transmits everything on all slave
                 interfaces.   This mode provides fault tolerance.

         802.3ad or 4

                 IEEE 802.3ad Dynamic link aggregation.   Creates
                 aggregation groups that share the same speed and
                 duplex settings.   Utilizes all slaves in the active
                 aggregator according to the 802.3ad specification.

                 Slave selection for outgoing traffic is done according
                 to the transmit hash policy, which may be changed from
                 the default simple XOR policy via the xmit_hash_policy
                 option, documented below.   Note that not all transmit
                 policies may be 802.3ad compliant, particularly in
                 regards to the packet mis-ordering requirements of
                 section 43.2.4 of the 802.3ad standard.   Differing
                 peer implementations will have varying tolerances for
                 noncompliance.

                 Prerequisites:

                 1. Ethtool support in the base drivers for retrieving
                 the speed and duplex of each slave.

                 2. A switch that supports IEEE 802.3ad Dynamic link
                 aggregation.

                 Most switches will require some type of configuration
                 to enable 802.3ad mode.

         balance-tlb or 5

                 Adaptive transmit load balancing: channel bonding that
                 does not require any special switch support.   The
                 outgoing traffic is distributed according to the
                 current load (computed relative to the speed) on each
                 slave.   Incoming traffic is received by the current
                 slave.   If the receiving slave fails, another slave
                 takes over the MAC address of the failed receiving
                 slave.

                 Prerequisite:

                 Ethtool support in the base drivers for retrieving the
                 speed of each slave.

         balance-alb or 6

                 Adaptive load balancing: includes balance-tlb plus
                 receive load balancing (rlb) for IPV4 traffic, and
                 does not require any special switch support.   The
                 receive load balancing is achieved by ARP negotiation.
                 The bonding driver intercepts the ARP Replies sent by
                 the local system on their way out and overwrites the
                 source hardware address with the unique hardware
                 address of one of the slaves in the bond such that
                 different peers use different hardware addresses for
                 the server.

                 Receive traffic from connections created by the server
                 is also balanced.   When the local system sends an ARP
                 Request the bonding driver copies and saves the peer's
                 IP information from the ARP packet.   When the ARP
                 Reply arrives from the peer, its hardware address is
                 retrieved and the bonding driver initiates an ARP
                 reply to this peer assigning it to one of the slaves
                 in the bond.   A problematic outcome of using ARP
                 negotiation for balancing is that each time that an
                 ARP request is broadcast it uses the hardware address
                 of the bond.   Hence, peers learn the hardware address
                 of the bond and the balancing of receive traffic
                 collapses to the current slave.   This is handled by
                 sending updates (ARP Replies) to all the peers with
                 their individually assigned hardware address such that
                 the traffic is redistributed.   Receive traffic is also
                 redistributed when a new slave is added to the bond
                 and when an inactive slave is re-activated.   The
                 receive load is distributed sequentially (round robin)