L11
Load balancing, redundancy and clusters
Cram this first — memory hook
Balance = share work; failover = take over; backup = go back.
Exam trap
Three web servers behind one unprotected load balancer can still have a single point of failure.
The slide facts
Load balancers select a back-end server and may keep a client on the same available server. Clusters can be active-passive or active-active. Reliability terms include Mean Time Between Failures , Mean Time to Failure and Mean Time to Repair .
If you have time — extra
A load balancer gives clients one front door and distributes work among multiple back-end servers. Health checks help avoid sending new requests to an unhealthy server. Session persistence keeps a client's related requests on the same server when needed. Active-passive clustering keeps a standby ready; active-active designs let multiple nodes serve work. Neither label alone guarantees that application state and data remain correct during failure. Redundant links, spare equipment and an Uninterruptible Power Supply () address different failure points. Backups preserve recoverable history and can help after deletion or ransomware; redundant live copies may replicate the same damage. Mean Time Between Failures () describes intervals between failures, Mean Time to Failure () describes time until failure, and Mean Time to Repair () describes repair duration. They are measurements/estimates, not promises for the next device.
Walkthrough
A request reaches the virtual service address. The balancer selects a healthy server. The server responds through the designed return path. If a node fails, health checks and failover reduce disruption; the balancer itself also needs redundancy.