Some of the coolest features/ enhancements in Hyper-V 2016

VM compute resiliency: This will help providing resiliency to transient issues like a temporary disconnection of a cluster node due to some network issues or if the cluster service itself on the node crashes etc. The VMs will still continue working "Unmonitored" even if the node falls out of cluster membership into an isolated state. Here the unmonitored state of the VM implies that it is no longer monitored by cluster service. The default resiliency period is 4 minutes. This means the Unmonitored VMs will be allowed to run on that isolated node for 4 minutes and after that VMs will be failed over to a suitable node/ nodes in the cluster. And that particular node which is isolated is moved to a down state. The cluster service itself is now not a necessary dependency for a VM to run. As long as connectivity exists the VM will continue working.

Node quarantine: If a cluster node is isolated certain number of times (default is 3) within an hour it will be moved to quarantine state and the VMs running on it (if any) will be failed over to another suitable node/ nodes in the cluster. 

Event 1 - cluster service stopped on node A - node A isolated (down) - cluster service restarted - node A online
Event 2 - cluster service stopped on node A - node A isolated (down) - cluster service restarted - node A online
Event 3 - cluster service stopped on node A - node A down - node A quarantined

The node will be quarantined for a period of 2 hours by default. But the administrator can manually start the cluster service on that node to join it back to the cluster.

VM storage resiliency: If there is a storage interruption, the VM identifies it and it will pause all the IO's for a certain duration and once the storage is available all IO operations will be resumed. This is very helpful in case of transient storage issues, saving the VM from blue screening or crashing. If the storage path is not back online after a certain period of time, it will pause the VM. Once storage comes back it auto resumes.

VM memory run time resize: You can now increase/ decrease RAM of a running VM.

Hot add/ remove VM network adapters: VM network adapters can also be added or removed on the fly.

Cluster OS rolling upgrades: With this feature you can upgrade your Hyper-V 2012 R2 cluster to Hyper-V 2016 cluster without shutting down the cluster. You can upgrade your existing cluster in 2 ways. Either you can add new 2016 nodes to the 2012 R2 cluster, migrate workload to new 2016 nodes and evict old nodes. Or you can evict one of the existing 2012 R2 node, do a clean installation of 2016, add it back to the cluster and do the same for rest of the nodes. Once all the nodes are 2016, you can update cluster functional level to 2016.

Storage Spaces Direct - Volumes and Resiliency

Storage Spaces Direct (S2D) is the Microsoft implementation of software defined storage (SDS). This article briefly explains about the different types of volumes that can be created on a S2D cluster. Once you enable S2D using Enable-ClusterS2D cmdlet, it will automatically claim all physical disks in the cluster and forms a storage pool. On top of this pool you can create multiple volumes which is explained below.

Mirror

• Recommended for workloads that have strict latency requirements or that need lots of mixed random IOPS
• Eg: SQL Server databases or performance-sensitive Hyper-V VMs
• If you have a 2 node cluster: Storage Spaces Direct will automatically use two-way mirroring for resiliency
• If your cluster has 3 nodes: it will automatically use three-way mirroring
• Three-way mirror can sustain two fault domain failures at same time

new-volume -friendlyname "Volume A" -filesystem CSVFS_ReFS -storagepoolfriendlyname S* -size 1TB

• You can create two-way mirror by mentioning "PhysicalDiskRedundancy 1"

new-volume -friendlyname "Volume A" -filesystem CSVFS_ReFS -storagepoolfriendlyname S* -size 1TB -PhysicalDiskRedundancy 1

Parity

• Recommended for workloads that write less frequently, such as data warehouses or "cold" storage, traditional file servers, VDI etc.
• For creating dual parity volumes min 4 nodes are required and can sustain two fault domain failures at same time

new-volume -friendlyname "Volume B" -filesystem CSVFS_ReFS -storagepoolfriendlyname S* -size 1TB -resiliencysettingname Parity

• You can create single parity volumes using the below

new-volume -friendlyname "Volume B" -filesystem CSVFS_ReFS -storagepoolfriendlyname S* -size 1TB -resiliencysettingname Parity -PhysicalDiskRedundancy 1

Mixed/ Tiered / Multi-Resilient (MRV)

• In Windows Server 2012 R2 Storage Spaces, when you create storage tiers you dedicated physical media devices. That means SSD for performance tier and HDD for capacity tier
• But in Windows Server 2016, tiers are differentiated not only by media types; it can include resiliency types too
• MRV = Three-way mirror + dual-parity
• In a MRV, three-way mirror portion is considered as performance tier and dual parity portion as capacity tier
• Recommended for workloads that write in large, sequential, such as archival or backup targets
• Writes land to mirror section of the volume and then it is gradually moved/ rotated in to parity portion later
• Each MRV by default will have 32 MB Write-back cache 
• ReFS starts rotating data into the parity portion at 60% utilization of the mirror portion and gradually as utilization increases the speed of data movement to parity portion also increases
• You should have min 4 nodes to create a MRV

new-volume -friendlyname "Volume C" -filesystem CSVFS_ReFS -storagepoolfriendlyname S* -storagetierfriendlynames Performance, Capacity -storagetiersizes 1TB, 9TB

References:

Microsoft

Blog cdemi