James O'Neill: Dynamic memory is not the same as the abilities found in VMware. Briefly, VMware can swap memory to disk at the virtualization layer. Just because it can, doesn’t mean it should: swapping to disk is best managed by the OS which can prioritize what should stay in memory and what shouldn’t. To minimize swapping, VMware has page sharing, which looks for pages duplicated between OS instances. I’ve dodged the word “over-committed” because strictly, over-committed means there is more memory in use than there is available – so paging is required, page sharing allows more memory to be allocated than is present without being overcommitted (in the strict sense of the term). Generally paging sharing works with homogenous OSes which are already memory constrained (i.e. when the largest proportion of memory is the core of the OS and only a small amount is unique data). When the VMs have a lot of memory there is a bigger overhead comparing the pages of each to see if there is a match, and if they have a lot of memory a lot of that will be used by data which varies between the VMs. Heterogeneous OSes will mean that there isn’t common code loaded to share – and in turn that means that instead of viewing the servers which host VMs as a cloud, where any VM can go to any server, to keep a good level of page sharing we need to keep all the VMs using one OS on one server and all those running another OS on another server....