Deploying Load-Balancing Network Topology to GNU/Linux

The first thing to do is to identify available network interfaces. To do that, simply:

The next thing is to have the bonding kernel module installed. Here is the command:

Then, check and load the module:

The next step is to create the bonding network interface. In this example, we name it (bond0). You can create this network interface in /etc/network/interfaces.d directory.

with the following:

NOTE:

1. For <connectionType>, you can use the following:
   • "static" for static IP configurations
   • "dhcp" for DHCP automated allocations (If you're unsure or you are a home-user, use this)
2. For load balancing <mode>, explained in detail in each subsections, you can use either of the following:
   • "balance-rr" for round-robin load balancing.
   • "balance-xor" for MAC addresses XOR matching selection.
   • "balance-tlb" for adaptive transmit load balancing (transmit is load-balanced using MII link only)
   • "balance-alb" for adaptive load balancing (receive is load-balanced using ARP negotiation only)
3. For "bond-miimon", it is periodic checking timing to inspect any link failure. One can set it to 100.
4. For "bond-downdelay", it is the wait timing to conclude a link is confirmed down. Set 0 to disable it.
5. For "bond-updelay", it is the wait timing to use a link when it is up. Set 0 to disable it.

One example would be:

This mode transmit and receive by rotating each slave links per packet sends. Hence, it uses all links in a round-robin manners.

The upside is that:

1. It offers fault tolerance (able to switch between links on-the-fly).
2. It offers load balancing across all links.
3. Suitable for consumer laptop, desktop, and server.
4. Suitable for commercial grade server.

This mode uses the primary slave for both transmit and receive. Unless it fails, the second and subsequent links take over its role.

The upside is that:

1. It offers fault tolerance (able to switch between links on-the-fly).
2. Suitable for consumer laptop, desktop, and server.

This mode transmit and receive by selecting link from the slave links pool through XOR source MAC address with destination MAC address.

The upside is that:

1. It offers fault tolerance (able to switch between links on-the-fly).
2. It offers load balancing across all links.
3. Suitable for local area network connections with multiple exit nodes (routers).

This mode sends transmit and receive across all interfaces.

The upside is that:

1. It offers fault tolerance (other interfaces can still work should one of the slave links went down).
2. High availability.

This mode is for IEEE 802.3ad compliant network.

The upside is that:

1. It offers fault tolerance (other interfaces can still work should one of the slave links went down).
2. High availability.
3. Autonomous configuration.
4. It offers load balancing.

This mode transmit and receive by balancing outgoing traffic across links based on their respective relative computed loads. The receive is the current active slave only. Should the current active slave failed, another slave takes over the receive job.

The upside is that:

1. It offers fault tolerance (able to switch between links on-the-fly).
2. It offers load balancing across all links.
3. Suitable for commercial grade server deployment (fixed hardware setup).

This mode transmit and receive by balancing both traffics across links using ARP negotiation. The bonding driver intercepts the ARP replies and overwrites the source hardware address with one of the slaves's unique hardware address.

The upside is that:

1. It offers fault tolerance (able to switch between links on-the-fly).
2. It offers load balancing across all links.
3. Suitable for commercial grade server deployment (fixed hardware setup).

In general, every slave links now must have the following statements:

1. bond-master can be self-defined (recommended bond0, bond1, ...). This is essentially creating a singular network interface.

NOTE: since we defined slaves inside bond0, there is no need to re-define bond-primary in every single interface.

Secondly, the iface definition must use manual instead. This is because the bonding driver will manage the connection/disconnection on its own autonomously via manual interfacing with slave links. Example:

Therefore, a common pattern would be:

Example:

Most OS will setup Ethernet link autonomously during installation and will append its settings inside /etc/network/interfaces. Your job is to extract the default settings out and create an independent /etc/network/interfaces.d/<name> config file instead. Here is an example of a working Ethernet configuration with the name "enp2s0".

Wifi link driver and interfaces usually do not exist as it is managed by different network application like wpa-* list of programs. It is no advisable to set manual wifi configurations (except your consumer laptop or pc is using terminal interface). For Wifi, let the appropriate wifi network program to handle the wireless connectivity.

If your load balancing mode is balance-tlb or balance-alb, you need to install and setup ethtool package. Otherwise, you may skip this step. To install:

Optionally, unless absolute necessary that you need bonding module to be loaded on boot, you simply add the following into /etc/modules:

Now that the setup is ready, you can restart networking service. Use root account to:

This will take some times. If there is any wrong configurations (e.g. accidentally use balance-tlb mode or balance-alb mode when many dependencies are not configured properly), this service restart will fail. You can use the following to read the error message:

To test Ethernet connection, you simply detach/attach the network cable to the port. You should experience uninterrupted connection when the cable is detached.

To test Wifi connection, you simply connect/disconnect your trusted network cable to the port. You should experience uninterrupted connection when the wifi is disconnected.

To check bonding status, you can read from the file inside /proc/net/bonding/<bond name>. Example:

To restart network as non-root user, you need to:

1. disable the primary slave.
2. restore all subsequent slaves connections (e.g. get wifi connected and stable).
3. enable the primary slave.