The host OS must have sufficient memory to provide services such
as I/O virtualization, VM snapshot, and management to support the
child partitions. The host reserves a minimal amount of memory, called
root reserve, which cannot be allocated to VMs. In general, this minimal
amount is often too low. This lack of memory invites the risk of the
VMs starving the host during periods of high activity, resulting in
the host OS running poorly and impacting the Microsoft Storage Spaces
and Hyper-V management functions.
Dell recommends
that you reserve at least 8 GB of memory for the host OS on each compute
node, by not allocating this memory to VMs.
Plan
your Converged solution so that when VMs failover to the remaining
nodes during a compute node failure, sufficient memory is available
to manage the additional VM load. Dell recommends that the amount
of memory specified for all VMs hosted by the compute cluster not
exceed the memory available for each node multiplied by the number
of nodes minus one. This recommendation maintains the guaranteed optimal
performance levels in the event of a single compute node failure.
This is shown in the following equation:
Total
amount of Memory available for VMs=(Memory available Per Node – 8
GB)*(Total # of nodes–1)
For example, in a
two-node Converged cluster with 128 GB of memory on each node, and
8 GB of RAM reserved for the host, the total memory available to be
allocated to all VMs in the solution should not exceed (128–8)*(2–1)
= 120 GB of memory. If each VM was configured to use 2 GB of RAM the
solution could support up to 60 total VMs, or 30 VMs per node.
In another example, a three-node Converged cluster
with 256 GB of memory on each node allows for 496 GB of memory that
can be allocated to all VMs hosted by the compute cluster. In this
example, if one node fails, then the 248 GB on each of the two remaining
nodes is sufficient to provide the allocated memory to VMs and still
have 8 GB in reserve for each of the node’s host OS.