A pre-experiment confirms that moving a fixed service MAC address (mac0) and service IP address (ip0) between nodes allows TCP connections to complete successfully. This validates a core design for an on-demand access system where a target machine stays in Suspend and is woken via Wake on LAN (WoL). Clients always connect to a stable endpoint, unaware of the backend's power state.

The test used Linux namespaces to simulate a standby node (A) holding the identity, a client (C) initiating a connection, and a target (B) that takes over after a handoff. By dropping packets instead of sending RST, A kept the connection attempt alive while B booted. This prevents clients from failing fast during the wake-up gap, a critical requirement for seamless user access.

Key operational requirements emerged: enforce single ownership of the identity to avoid ARP conflicts, use DROP semantics to enable client retries, and trigger gratuitous ARP (GARP) for fast L2/L3 convergence post-handoff. Controlling ARP flux via specific sysctls is also baseline. These findings provide a blueprint for moving from simulation to a production power-saving architecture for servers or workstations.