6.0.1

n
To resolve a CPU resources problem, set an explicit CPU reservation for the Primary VM at an MHz
value sufficient to run its workload at the desired performance level. This reservation is applied to both
the Primary and Secondary VMs, ensuring that both VMs can execute at a specified rate. For guidance
in setting this reservation, view the performance graphs of the virtual machine (before Fault Tolerance
was enabled) to see how many CPU resources it used under normal conditions.
Increased Network Latency Observed in FT Virtual Machines
If your FT network is not optimally configured, you might experience latency problems with the FT VMs.
Problem
FT VMs might see a variable increase in packet latency (on the order of milliseconds). Applications that
demand very low network packet latency or jitter (for example, certain real-time applications) might see a
degradation in performance.
Cause
Some increase in network latency is expected overhead for Fault Tolerance, but certain factors can add to
this latency. For example, if the FT network is on a particularly high latency link, this latency is passed on to
the applications. Also, if the FT network has insufficient bandwidth (fewer than 10 Gbps), greater latency
might occur.
Solution
Verify that the FT network has sufficient bandwidth (10 Gbps or more) and uses a low latency link between
the Primary VM and Secondary VM. These precautions do not eliminate network latency, but minimize its
potential impact.
Some Hosts Are Overloaded with FT Virtual Machines
You might encounter performance problems if your cluster's hosts have an imbalanced distribution of FT
VMs.
Problem
Some hosts in the cluster might become overloaded with FT VMs, while other hosts might have unused
resources.
Cause
vSphere DRS does not load balance FT VMs (unless they are using legacy FT). This limitation might result in
a cluster where hosts are unevenly distributed with FT VMs.
Solution
Manually rebalance the FT VMs across the cluster by using vSphere vMotion. Generally, the fewer FT VMs
that are on a host, the better they perform, due to reduced contention for FT network bandwidth and CPU
resources.
Losing Access to FT Metadata Datastore
Access to the Fault Tolerance metadata datastore is essential for the proper functioning of an FT VM. Loss of
this access can cause a variety of problems.
Problem
These problems include the following:
n
FT can terminate unexpectedly.
Chapter 2 Troubleshooting Virtual Machines
VMware, Inc. 15