You can reduce the chances of poor performance and data loss with a solid vSphere cluster design, one that keeps networking and configuration consistency in mind.
VMware vSphere clusters address the risks inherent to server virtualization, such as VM loss and finite host compute resources. With proper cluster design, you can mitigate these risks.
When you design vSphere clusters to run a population of VMs, you manage these clusters through vCenter. You can use vCenter to set the policy for how the cluster should operate, and the cluster automatically implements your policy.
In the 13 years since VMware Infrastructure 3's release, vSphere clusters have acquired many features, but the two core features remain High Availability (HA) and the Distributed Resource Scheduler (DRS). These features protect VMs from data loss and optimize VM performance.
Navigate vSphere cluster design
On a vSphere cluster, every VM should be able to run on any physical host and move freely between hosts. To achieve this VM mobility within the cluster, your vSphere hosts must have consistent hardware and configuration. The most effective vSphere clusters have interchangeable and identical hosts.
To achieve this VM mobility within the cluster, your vSphere hosts must have consistent hardware and configuration.
A vSphere cluster is made up of physical vSphere hosts with the same CPUs, similar amounts of RAM, and the same number and type of storage and network adapters. Most often, you would purchase the entire cluster from a single vendor at the same time.
Use HA and DRS
VMware vSphere HA rapidly and automatically restarts VMs running on a physical vSphere host that fails. When you consolidate dozens of servers inside VMs onto a single physical server, you make that server a point of risk. HA protects you against that risk by powering on VMs using the remaining hosts when a specific host fails.
You configure, manage and monitor HA through vSphere. But vSphere hosts manage the actual failover, and HA should protect the vCenter VM.
DRS reduces performance issues on a vSphere host. Dozens of servers share the vSphere host's physical resources. If multiple VMs are busy at the same time, then VMs can saturate host resources, and application performance suffers.
A DRS cluster balances VMs across multiple physical hosts to minimize the chance that one host becomes overloaded while another host has spare resources. DRS uses vMotion to move running VMs from host to host, which minimizes the risk of a VM lacking necessary resources. DRS runs entirely on vCenter, where you can initiate vMotion via the load-balancing calculations.
You can set a DRS cluster to only make recommendations. However, a fully automated cluster that immediately takes vMotion recommendations better avoids potential performance issues.
Maintain consistency
All vSphere hosts in a cluster should contain the resources used by each VM. When you design a vSphere cluster, VM port groups should have the same name, virtual LAN (VLAN) ID, security policy and physical network interface card speed across all cluster members. All vSphere hosts should have access to the data stores containing the VMs, regardless of whether they are Fibre Channel, Network File System, iSCSI or vSAN data stores.
The vSphere host must also have consistent networking in order for DRS to use vMotion. You must enable vMotion on all vSphere hosts with the same name, VLAN ID and TCP/IP subnet on VMkernel network ports.
Inconsistency in a vSphere cluster design leads to clusters that don't work as expected. To make and maintain a consistent cluster, document every step in your build and configuration process, and follow the document -- although documentation alone doesn't achieve consistency.
An automated build process delivers more consistent clusters, so consider scripting your configuration. However, build scripts only enforce consistency at one point in time, and vSphere host configuration can change over time, which leads to future inconsistencies.
You can also use VMware Host Profiles, a vCenter feature that defines the desired host configuration for a cluster and automates the process of making each host match that configuration. With Host Profiles, you achieve a consistent configuration, can audit for consistency over time and can force hosts to return to a consistent configuration.
