Figure 3.34 The SLB implementation of HNVv2
The SLB MUX is the SLB Multiplexer, which is implemented as its own VM. Its role is
to handle the initial request from the client. Note that it is possible to have multiple
SLB MUXs, which would be desirable in order to remove any single point of failure.
The multiple SLB MUXs offer the same VIP, and this is possible through the use of
Borderless Gateway Protocol (BGP). Each VIP on the SLB MUX becomes a /32 route.
(/32 is the number of bits in the subnet mask, and because 32 indicates all possible
bits, it means that it’s a route for a specific IP address, that of the VIP.) The VIP /32
route is advertised up to the physical switches via BGP from the SLB MUXs that host
the VIPs. The switches receive the advertisement of the routes and now are aware that
the switch can get to the VIP through any of the MUXs. Because all the routes are
equal, the switch will use a protocol called Equal Cost Multi-Pathing (ECMP) to
distribute the requests evenly between all of the MUXs offering the same VIP. This
enables scalability and the ability to easily add and remove MUXs from the
deployment. BGP also has a keep-alive capability that implements constant
communication between the MUX and the rest of the network infrastructure. If a
MUX goes down, the physical switches will become aware that it is not available and
stop sending traffic by updating its own routing tables.
The SLB MUX accepts the incoming packet, which is addressed to the VIP. However,
the actual backend VMs have dynamic IPs (also called datacenter IPs, or DIPs), which
are the internal IP addresses they use to communicate on the datacenter network. On
most load balancers, the packet has NAT performed to rewrite the destination of the