The basic cause of host-level swapping is memory overcommitment from running memory-intensive VMs whose combined configured memory is greater than the amount of host physical memory available.

To resolve this problem, consider the following factors:

• Reduce the level of memory overcommitment (for example, reduce oversized VMs and add more memory where necessary, add more hosts to the cluster if necessary, and so on).
• Enable the balloon driver in all the VMs by installing VMware Tools.
• Reduce memory reservations. This will be covered in Chapter 5, vSphere Cluster Design.
• Use resource controls (Shares, Reservations, and Limits) with careful understanding and planning to reduce memory contention on the most critical VMs.

Note that this will not reduce the total memory contention on this host, but it will allow you to mitigate the risk to the performance of your most important VMs.