Purely Cloud Storage Area Network (SAN) includes high availability mechanisms that are familiar if you are like the 80% of the array market that uses dual-controller array approaches. It optimizes data safety, but there might be a few seconds of VM inactivity during component failover. If a host substantially fails with Hyper-V HA, a VM will be restarted on another host, and the data should be moved to the new server as soon as possible over the Arista Networks 10 Gigabit backbone. Storage and network layout is optimized for aggressive streaming throughput to hosts, not small random reads. But it doesn’t stop there.
HA standpoint; there’s no single point of failure.
Drive failure: data is reconstructed from RAID6 dual parity. Multiple drives can fail and the system can keep going.
Link or switch failure: each of the 2 controllers has 2 10Gb ports. A failed link will switch the active connection to a different link.
Controller failure: The controllers are active/passive. If a controller fails, the other will take over. All writes are to a controller’s NVRAM, but they are mirrored synchronously to the passive controller, so nothing is lost if a write has been committed.
Other component failures: all other parts (e.g. power supplies) are redundant.
Host HA Maintenance mode. If a given Hyper-V host needs to enter maintenance mode for Hyper-V patching, data for the evacuated VMs will moved to another host from the SAN.
Host failure. Here, VMs are restarted by Hyper-V HA on alternative hosts on the cluster after a failure. The expectation is seconds-to-minutes of restart time. During VM restart, the storage’s design focus is on optimizing the startup on the new host.