What is HCI Mesh in vSAN 7 Update 2?

VMware released with vSAN 7 Update 2 a new feature called HCI Mesh to provide better scaling capabilities in a vSAN environment.

Before vSAN 7 Update 2, there’s always been a very tight integration between vSphere and vSAN that made both products dependent from each other, and it wasn’t possible either to consume storage from one vSAN cluster to another.

Now things have changed with HCI Mesh as a non-vSAN Cluster can now take advantage of a remote vSAN cluster to start provisioning VMs to a remote vSAN datastore.


Use Cases

Why would we use HCI Mesh and how can that feature help us in a vSphere/vSAN environment?

Hardware Maintenance

HCI Mesh can be very helpful to move data while planning any type of maintenance windows, to apply a patch or upgrade your hardware.

Balance Storage capacity across Cluster

To help you balance storage capacity across multiple clusters, you can use HCI Mesh to simply move data to the vSAN cluster of your choice based on capacity usage.

Storage as a service

Cloud providers can leverage HCI Mesh to provide storage to their customers so that they can consume storage on demand.

License Optimization

With HCI Mesh, we can now separate compute and storage and provision license accordingly to reduce cost.

Scale compute resources without adding storage

To scale out a traditional vSAN cluster to provide more compute resources, it is necessary to provision a new host by adding both compute and storage capacity.

This results often in an unbalanced compute/storage ratio with over-provisioned storage capacity being wasted. With HCI Mesh, it is now possible to better scale compute and storage independently from each other.


HCI Mesh Topologies

There are two different topologies that can be used based on your requirement.

  • Cross-Cluster Topology
  • Storage-only Cluster Topology

Cross-Cluster Topology

In a Cross-Cluster topology, the following features and use cases can be addressed:

• Up to 5 vSAN remote datastores can be mounted

• Mounting a remote datastore can be bidirectional

• VMs can run on both client and server clusters

• Can be very useful to starving clusters requiring storage capacity

Cross-Cluster Topology

Storage-only Cluster Topology

In this topology, a vSAN cluster acts as a storage provider to only provide storage to other client clusters. Please put in mind, as a best practice that the storage cluster does NOT run any workloads.

The Storage-Only cluster topology can be used in these use cases:

• Best to properly and independently scale compute and storage resources

• Ideal topology to provide storage to disk-less client clusters

• Best option to save license costs as no workload run on the storage-only cluster

storage-only topology


HCI Mesh: Network requirements and recommendations

Network connectivity requirements:

Both layer 2 and layer 3 are supported for intercluster connectivity.

Network redundancy recommendations:

• Use dual top-of-rack (ToR) switches on the rack.

• A leaf-spine topology is best for core redundancy but also reduces latency.

• Use multiple NICs on each ESXi host.

• A single vSAN VMKernel is required without air-gap support.

Network performance requirements:

HCI Mesh has the same latency and bandwidth requirements as a regular local vSAN cluster.


HCI Mesh Licensing

Licensing requirements:

• vSAN Enterprise edition is required on each cluster participating in a mesh topology.

Scalability limits: because multiple vSAN clusters can participate in a HCI mesh topology, the scalability limitations mostly apply to the vSAN datastore objects:

• A single vSAN datastore can be mounted on a maximum of 64 hosts, which include both server and client hosts.

• A client cluster can mount a maximum of five remote vSAN datastores.

• A server cluster can export its datastore to a maximum of five client clusters.

Once again, VMware released here a very nice feature that will be for sure very useful to many of us. Hope you like the blog post and do no hesitate to share.

Leave a Reply

Your email address will not be published.