Load Balancing: Virtual Servers, Services, Monitors, and Load Balancing Methods
The object chain
The exam tests whether you can trace a request from an LB vserver to a service and explain why a service is eligible. An LB vserver has a type (HTTP, SSL, TCP, UDP, ANY) that controls which features and expressions are valid. A service is an IP address, port, and protocol; a service group holds members. Monitors bind to services or service-group members, not directly to the LB vserver. If no monitor is bound, the appliance generally treats the service as up because it has no health signal to contradict the enabled state.
Choosing a method
- Round Robin: sequential selection; weights change the proportion of selections.
- Least Connection: fewest active connections, with weights considered.
- Least Response Time: combines connection count and observed response time.
- Least Bandwidth and Least Packets: select by measured throughput or packet rate.
- Source IP Hash, Destination IP Hash, Source-Destination Hash: deterministic hashing for stickiness-like distribution.
- Custom Load: uses a load value, often supplied by a custom monitor or SNMP.
Weights are relative, not capacity guarantees. A monitor's interval and retries determine detection time: a service is marked down after the configured number of consecutive failed probes. Down state flush and graceful disabling matter when draining a server.
Worked example
Three services have weights 3, 1, and 1 under Round Robin. Normalize the weights to 3:1:1, so five consecutive selections are A, A, A, B, C, then the cycle repeats. If B fails its monitor, B is removed from selection and the remaining A:C ratio becomes 3:1; the ADC does not renormalize B's weight onto A and C as if B had never existed. If the same services used Least Connection, A would receive more new connections only while its active count is lower relative to its weight; a busy A can lose the next selection even though its weight is highest.
Traps
Binding a monitor to the LB vserver does not health-check the services. A service with no monitor is not automatically down. show service distinguishes administrative state from effective state; a service can be enabled but effectively down because a monitor failed. Least Connection counts active connections, not request rate. Source IP hash is not a persistence timeout mechanism, and it collapses many clients behind one NAT address onto one service.