I always recommend to create a dedicated management cluster for your vSphere virtual environment, but what is a dedicated management cluster, and why is it so important to have one? Not only is it best practice, there are real reasons why you should choose to do this.
What is a Dedicated Management Cluster?
In it’s most basic form, it is assigning two or three hosts to a cluster, specifically for the management of the rest of your virtual environment. Running on these vSphere hosts, you would have management virtual machines, such as vCenter, VADP, Update Manager, vShield Manager, VSM, vRealize / vC Ops, vCloud Director, SRM, and other non VMware systems such as DNS, Active Directory, Endpoint Protection, SAN and network management tools, backup software controllers, etc..
Ideally, in best practice, you would dedicate three hosts in a HA cluster, although most commonly I see two hosts in a HA cluster. You can also have a single host dedicated to management tasks, although this does reduce availability from what you can achieve with two or three hosts.
The management cluster should be the first hosts you power on in your data centre after a power outage or problem, and if you have an issue with vCenter and need to manually power it up through the command line, you have less hosts to search through to find the registered VM.
By separating management of the virtual environment from the operation of the virtual environment, you can be assured that the resources available for management tasks are both fully available to you when you need them, and also do not consume resources that would affect the rest of your production environment (think of an incident where you need to administer VMs that are under heavy strain).
What do you not need?
Something that is not often mentioned on other blogs, or by other consultants when they recommend the best practice of a dedicated management cluster, is what is not required.
You don’t need DRS – the purpose of DRS is to allow the balancing of VM workloads in the event that there is an over-consumption of resources that would affect other VMs on the host. As management tasks are relatively low-intensity and less demanding (and you may have only two or three hosts), the additional overhead of DRS – and the license cost of Enterprise or Enterprise Plus.
You don’t need to use Fibre Channel – if you have an FC connected SAN, you can save on FC cards and expensive fabric licensing and SAN storage – instead you can use an iSCSI or NFS connected SAN or NAS, or (considering the low number of hosts) a SAS connected shared storage SAN. These can offer excellent performance for a low number of hosts with low throughput.
You don’t need to use Distributed Switch – again, with the low number of hosts, and to reduce on Enterprise Plus licensing costs, there are few benefits of the vDS that you actually need for a management cluster.
You don’t need fast, large, or modern hosts – this might be a bit contentious for some, but I often give general advice to use a previous generation of hosts for the management cluster. That is, if a company is purchasing new hosts with more RAM and newer CPUs, and their old servers are still reliable and serviceable, then re-use the older hosts for the management cluster. To size the hosts for the management cluster, add up all the RAM required for all the management VMs (this can be achieved by looking at what they are consuming in your current environment), and ensuring that your smallest host can run all the management VMs in an HA event. The CPU load of management VMs is often low, and providing you have enough cores, the speed of hardware is most often not a factor.
Obviously, your mileage may vary, but I give general advice to distribute the hosts into different racks (or better yet, different rows). I also recommend multiple network uplinks to the hosts, and storage – if using iSCSI or NFS.
Although VCSA (vCenter Server Appliance) has an embedded database, and will therefore be on the management cluster, you need to consider the availability, costs and performance requirements for any other database server on your management cluster. You may already have a clustered database solution (or choose to leverage HA), or your licensing costs for the database solution (based on cores or CPUs) may exclude purchasing a license for relatively low load. This is something that will vary for each company.
Although you can mix and match hardware, vendors and components, the focus should be on reliable systems. Don’t re-awaken an old bit of junk that takes 15 minutes to power up!