- Host/Node Setup
- Make sure the host and all nodes in a cluster have the BIOS Settings identical (All settings)
- Leave ~1GB to 1.5GB physical RAM to the host
- We leave 1.5GB of space on a dedicated to VM LUN
- We leave ~50GB to ~100GB of free space on a shared LUN/Partition
- We set the MiniDump option and lock the swap file down to 840MB
- wmic RECOVEROS set DebugInfoType = 3
- Always set up a standalone host with two partitions: OS and Data
- Hyper-V lays out a file equivalent in size to the vRAM assigned to VMs. We must have space for them.
- Snapshots/CheckPoints create differencing disks. These _grow_ over time.
- Deleting Snapshots/CheckPoints requires enough free space to create an entirely new Parent VHDX.
- vRAM assigned to the VM should not traverse NUMA nodes (performance) (more on hardware).
- vCPUs = Threads in CPU and must be processed in parallel thus # physical cores - 1 is best.
- GHz is preferred over CPU Core counts for most workloads.
- Be aware of the IOPS required to run _all_ workloads on the host/nodes.
- More smaller sized spindles is better than less larger size spindles = More IOPS.
- 10GbE should be the minimum bandwidth considered for any iSCSI deployments.
- At least _two_ 10GbE switches are mandatory for the storage path
- Broadcom physical NIC ports must always have VMQ turned off (blog post)
- We prefer to use Intel Gigabit and 10Gb Ethernet Server Adapters
- We start with a minumum of 4 physical ports in our hosts/nodes
- UPS Systems
- UPS Systems should have at least 1-1.5 Hours of runtime.
- Host/Nodes and storage should be tested for shutdown durations.
There are quite a lot of these types of posts on our blog. Please click through the category tags to find more!
Thanks for reading.
Microsoft High Availability MVP
Co-Author: SBS 2008 Blueprint Book