In a vSphere 6.5 cluster of 6 nodes, memory usage is very high on host1 and host2, and network bandwidth usage is very high on host3 and host4. The administrator needs to place host1 into maintenance

vSphere 6.5 introduced network utilization as a DRS load balancing metric. When host1 is placed into maintenance mode and DRS selects migration targets, it evaluates the current network bandwidth consumption on all remaining hosts and will steer VMs away from host3 and host4, which are already network-saturated, to avoid worsening their congestion.

The VM Distribution setting spreads VMs evenly by count across hosts but does not factor in the specific resource saturation states of host3 and host4 when choosing migration targets.

Disabling the memory metric causes DRS to rely primarily on CPU utilization for balancing decisions, which would ignore both memory and network conditions and could still place VMs on the saturated hosts.

Host affinity rules constrain where specific VMs may run but do not by themselves drive or prevent DRS from relocating VMs during a maintenance mode evacuation.