Linux bonding combines interfaces into a logical interface for redundancy or load sharing. It is not exclusive to CentOS and does not automatically move an IP between servers. Identify the failures to tolerate before selecting a mode.

Compare common modes

mode Name Main behavior
0 balance-rr Packet round robin; assess switch requirements and reordering
1 active-backup One active link with failover
2 balance-xor Hash-based interface selection requiring network planning
3 broadcast Sends through every member; not aggregate throughput
4 802.3ad Dynamic aggregation requiring matching switch LACP support
5 balance-tlb Adaptive transmit load balancing
6 balance-alb Adds handling for particular receive-load scenarios

802.3ad is not simple primary/backup operation. Calling alb network-card virtualization is also imprecise. Aggregate bandwidth does not mean one connection can use every member’s bandwidth at once; hashing and traffic distribution matter.

Check both ends before configuration

Confirm speed, duplex, switch ports, VLAN and MTU. Layer 3 settings such as IP and gateway usually belong on the bond, without conflicting IPs on members. Connecting across switches depends on support for the intended aggregation or failover topology.

CentOS 6 ifcfg and modprobe configuration do not replace management through NetworkManager in later environments. Avoid multiple tools modifying one interface. Prepare console access and rollback before remote changes so a lost connection remains recoverable.

miimon detects physical link status, but an up interface does not establish working upstream routes or services. Select link or ARP monitoring according to the topology, and inspect members and mode in /proc/net/bonding/bond0.

Test disconnecting the active link, switch-port failures, restoration and continuous business requests. Record failover duration and packet loss. One successful switch does not prove the same detector recognizes every failure.

References


blog comments powered by Disqus
Fork me on GitHub