Table of Contents
Advertisement
sufficient for systems that only run a single workload, such as a benchmarking
configuration, which does not change over the course of the system's uptime.
However, initial placement is not sophisticated enough to guarantee good
performance and fairness for a datacenter-class system that is expected to
support changing workloads with an uptime measured in months or years.
To understand the weaknesses of an initial-placement-only system, consider the
following example:
An administrator starts four VMs. The system places two of them on the first
node and two on the second node. Now, consider what happens if both VMs
on the second node are stopped, or if they simply become idle. The system is
then completely imbalanced, with the entire load placed on the first node.
Even if the system allows one of the remaining VMs to run remotely on the
second node, it will suffer a serious performance penalty because all of its
memory will remain on its original node.
Dynamic load balancing and page migration
To overcome the weaknesses of initial-placement-only systems, as described in
the previous section, VMware ESX combines the traditional initial placement
approach with a dynamic rebalancing algorithm. Periodically (every two seconds
by default), the system examines the loads of the various nodes and determines
whether it should rebalance the load by moving a virtual machine from one node
to another. This calculation takes into account the relative priority of each virtual
machine to guarantee that performance is not compromised for the sake of
fairness.
The rebalancer selects an appropriate VM and changes its home node to the
least-loaded node. When possible, the rebalancer attempts to move a VM that
already has some memory located on the destination node. From that point on,
the VM allocates memory on its new home node, unless it is moved again. It only
runs on processors within the new home node.
Rebalancing is an effective solution to maintain fairness and ensure that all
nodes are fully utilized. However, the rebalancer might have to move a VM to a
node on which it has allocated little or no memory. In this case, the VM can incur
a performance penalty associated with a large number of remote memory
accesses. VMware ESX can eliminate this penalty by transparently migrating
memory from the virtual machine's original node to its new home node. The
system selects a page, 4 KB of contiguous memory, on the original node and
copies its data to a page in the destination node. The system uses the VM
monitor layer and the processor's memory management hardware to seamlessly
remap the VM's view of memory, so that it uses the page on the destination node
for all further references, eliminating the penalty of remote memory access.
Chapter 2. Product positioning
69
Advertisement
Table of Contents
