Storage DRS is a feature of vSphere 5.0 onwards which aims to reduce the complexity of managing storage capacity and performance for virtual machines running on traditional shared storage.
SDRS helps simplify management of Datastores (LUNs / NFS Mounts) by analyzing the available datastores capacity and placing new virtual machines in datastores with the required available capacity. This is called "Initial Placement". SDRS also allows administrators to create rules ("Affinity" & "Anti-Affinity")
to ensure specific what VMs should be kept together or apart for performance / capacity reasons and finally, monitor I/O metrics and relocate VMs to different datastores where contention or capacity constraints exist.
As it is recommended to have one Large NFS Datastore in a Nutanix environment, the issue of initial placement (which SDRS helps resolve) is natively eliminated as all Virtual machines, regardless of performance or capacity requirements can and should be placed into one NFS Datastore which spans the entire Nutanix Cluster (and potentially multiple vSphere Clusters) and has access to all the available storage in the Nutanix Storage Pool.
So the vSphere administrator, or provisioning tool can simply place all Virtual machines into a single NFS Datastore, thus eliminating the requirement for the "Initial Placement" feature.
Next lets discuss, "Affinity" & "Anti-Affinity" Rules
which help optimize VM performance. As Nutanix runs a Virtual storage Controller (or CVM) on every node, Virtual machines running on different ESXi hosts get access to their own Storage controller.
As a result, there is no need for "Affinity" & "Anti-Affinity"
Rules at the SDRS layer, as these can be configured at the vSphere DRS layer. All Virtual machines running Business critical applications (vBCAs) with high I/O requirements should be configured with a "Virtual Machines to Hosts" rule with "Should run on this host" configured as you would in a traditional vSphere environment on centralized shared storage.
This ensures the Virtual machine does not move around automatically via DRS unless there is a HA or Maintenance event which necessitates a vMotion.
The added advantage of these rules in the Nutanix environment is the vBCAs have access to a virtual storage controller on the same ESXi host, thus eliminating the latency of accessing traditional centralized storage over a Storage Area Network (FC or IP) & improved performance with Data Locality & by Nutanix natively mitigating against the "Noisy Neighbor" issue.
Next we can talk about SDRS's Load Balancing based on the I/O Metric
which uses an I/O injector
to test the performance of a Datastore and the "I/O latency threshold
" to determine if Virtual machines should be relocated.
As Nutanix is designed around a VM centric model, as opposed to a per LUN or Datastore model, testing the performance of a Datastore is no longer relevant as its performance may change depending on the workload of a single node. Instead Nutanix recommends to use DRS anti-affinity rules for vBCAs with high I/O at the compute layer, to separate this workloads across the cluster to ensure both optimal compute and storage performance.
Load Balancing based on Space Utilization is only relevant where multiple Datastores are configured, each with a finite amount of storage. In the Nutanix world, one Datastore has access to ALL available storage within the Storage Pool thus eliminating the requirement to load balance based on Space utilization and the associated complexity around capacity management.
This also ensures optimal utilization of available capacity, ie: No silos of wasted storage!
Finally, what about SDRS "Datastore Maintenance mode"?
Again as Nutanix has one large Datastore which is actively services by all Nutanix nodes in the cluster, in the event you need to remove a node from the cluster, this operation is handled in the background by the Nutanix Distributed File System (NDFS) completely transparent to the vSphere cluster.
The process is simple, Place the ESXi host into maintenance mode, all VMs are automatically evacuated by vMotion, then shutdown the host. Then the Nutanix node can be removed from the cluster without downtime and with minimal impact to the cluster as the work of removing the node is distributed across all nodes in the cluster.
As a result, while SDRS is a great feature for mitigating the storage capacity and performance challenges in traditional SAN/NAS environment, SDRS is not required or advantagous in a Nutanix environment.
The Recommended configuration for Storage DRS is as follows;