Objective 7.1 – Configure VMware clusters
Create new cluster
Clusters require a datacenter object.
Add ESX/ESXi hosts to a cluster
Standalone resource pools are grafted into a cluster when the option is specified; not the default. The default is to drop the resource pools.
Removing a host from a cluster requires the host to be placed in maintenance mode i.e. all virtual machines exvacuated from the host.
Configure High availability basic/advanced settings
- Shared storage
- Identical networking configuration
- vMotioning a virtual machine between hosts is a good test. If vMotion works then HA will work too.
vSphere 4 sends a heartbeat every 15 seconds according to the VMware KB1006421
Host isolation response
This is what happens to virtual machines should the ESX/ESXi host become isolated from its peers; the default response is to shutdown the virtual machines. In order to stop this creating downtime consider a redundant management link i.e. a second service console or vmkernel interface; this portgroup should not have a default gateway. Alternatively add a second network adapter to the virtual switch. The other consideration is to configure a cluster-wide setting of ‘leaving virtual machines powered on’.
HA is configured on the cluster object.
When planning ESX/ESXi maintenance consider disabling HA by deselecting the ‘Enable Host Monitoring’ checkbox.
Enable/Configure VM Monitoring
HA can be configured to reboot a virtual machine if it becomes frozen for 30 seconds when the failure interval is configure as high, 60 seconds when the failure interval is configured as medium or 120 seconds when the failure interval is configured as low. The reset periods are 1 hour, 24 hours and 7 days for high, medium and low respectively.
Configuring DRS basic/advanced settings
DRS uses statistics collected by vCenter from the hosts; these statistics are collected every 20 seconds.
VM placement is only available if partial or fully automated automation levels are selected.
Virtual machines will only be moved around in the fully automated automation level. If manual or partial is selected then the administrator will be given recommendations. The general rule of thumb for migration thresholds is low utilised environment; aggresive threshold, high utilised environment; conservative threshold.
Distributed Power Management (DPM) uses the statistics collected to determine if it can vMotion virtual machines to another host so it can place that host in maintenance mode. Before placing the host in standby vCenter uses 40 minutes of statistics to validate the decision; before waking up a host vCenter uses 5 minutes of statistics.
DPM will target 63% utilisation across all hosts in the cluster.
DPM requires: vMotion and DRS and can leverage IPMI, iLO and WOL to wake up the hosts.
EVC allows you to mask out CPU instruction sets and features at the cluster level rather than a per virtual machine level. EVC scans the available processors in the cluster hosts to find a common baseline.
- AMD operton generation 1, RevE or later and Intel Core 2 or later.
- The hosts must be managed by vCenter.
- AMD-V or Intel-VT enabled.
- Memory execution prevention enabled
- The guest operating systems should honour / respect CPU ID masking.
Configure swap file
By default the swap file is placed with the virtual machine. If you manage the swap file via the cluster settings you can specify local storage for the swap file without breaking vMotion. Note though that vMotions will take longer as the vMotion would have to move the swap file and memory bitmaps.
HA host failure capacity requirements
There can only be a maximum of five primary hosts per cluster which dictates a maximum of four failover hosts. The primary hosts manage virtual machine inventories, heartbeats, virtual machine restarts and promote secondary hosts to primaries.
Analyse HA admission control
Admission control guarantees enough unused capacity is available in the cluster to satisfy the host failover capacity.
Use cases for DRS automation levels and migration thresholds
Manual – the administrator wants to manage everything.
Partial – the administrator wants to control when and where virtual machines are moved.
Fully – all migrations are automated using the migration threshold. The calculation looks at the target host load standard deviation (the DRS migration threshold), the current host load standard deviation (average resource usage across all hosts in the cluster) and a standard deviation (how much variation exists from the average).
Use cases for DPM policies
Off – No DPM actions i.e. nothing gets put into standby.
Manual – recommendations based on the calculated load.
Automatic – will put hosts into standby or out of standby depending on the calculated load.