Database Reference
In-Depth Information
TPS won't stop paging and swapping from happening; it actually allows them to happen.
What it will do is allow the paging and swapping to happen with the least possible
amount of performance impact.
Recognizing the fact that many of the virtual machines running may have identical sets of
memory content, TPS invokes the hypervisor to identify those duplicate contents of
memory and allows them to be shared. It does that by keeping a single read-only copy,
and it uses a copy-on-write mechanism to ensure the security and integrity of the data.
If you had three guest VMs running the Windows 8 operating system, the hypervisor
would only need to keep one copy of the operating system in memory for all three VMs
to share. The memory would be reclaimed and made available to the hypervisor, which
would give it back to the guest virtual machines to take advantage of on the physical
host.
In this example, one copy of the OS would be placed in memory. The other two virtual
machines would have pointers to the spot.
Tip
Transparent page sharing is more effective the more similar the VMs are. When
possible, put like operating systems on the same physical host.
Tip
A great resource for better understanding transparent page sharing in more detail,
as well as other memory management techniques, is the VMware performance
study titled “Understanding Memory Resource Management in VMware vSphere
5.0,” found at
http://www.vmware.com/resources/techresources/10206
.
Internet Myth: Disable Memory TPS
There is a myth circulating that TPS is an expensive process in terms of overhead, the
consequences of using it far outweigh the benefits, and it should be disabled.
In
Figure 7.5
,
you can measure the impact of having transparent paging disabled
compared to having it enabled. You see that enabling page sharing introduces a
negligible performance overhead when run in the default setting. When
Mem.ShareScanTime is set to 10 minutes for all workloads, there is still less than 1%
percent overhead experienced.
