With its latest vSAN version, VMware has tried to solve one of the biggest knocks on hyper-converged infrastructure: independent scaling of compute and storage, without requiring customers to buy additional nodes.
VMware vSAN software-defined storage is bundled in the virtualization vendor's vSphere cloud-computing stack. This week, VMware introduced vSphere 7 update 2. Of interest to enterprise storage customers is an enhancement to the HCI Mesh technology that VMware introduced last year.
The first iteration of VMware HCI Mesh allowed customers to share storage across vSAN HCI compute clusters. For example, a vSAN-enabled Oracle database cluster is able to securely share its unused storage with a vSAN-enabled SQL cluster. With vSAN 7 update 2, HCI Mesh allows compute-only nodes -- servers not configured as part of a vSAN deployment -- to consume storage from remote clusters, using native vSAN protocols.
VMware isn't ploughing new ground here. Nutanix, VMware's chief rival in the HCI market, allows customers to add compute-only or storage-only nodes that run its AHV hypervisor.
Dell EMC VxRail offers a similar HCI topology for scaling resources out and up. Dell Technologies became the majority stakeholder in VMware after buying EMC Corp. in 2015. (Dell also sells Nutanix HCI under an OEM deal. The Dell EMC XC Series packages the Nutanix stack on PowerEdge servers.)
What VMware wants to do is help ease customers into HCI who have been reluctant to deploy it, said Lee Caswell, a VMware vice president. He said data centers can add HCI Mesh compute-only clusters to existing vSAN clusters without an additional VMware license.
Sharing VMware vSAN storage using HCI Mesh, compute clusters now can consume unused capacity from mounted data stores that are attached to a vSAN storage cluster. Enterprises manage the compute-only clusters through vSphere, similar to the manner of virtual machines on a storage array.
"You don't have to pay anything special [to use HCI Mesh]. The full capacity and performance of your vSAN cluster is now accessible by remote compute clusters," Caswell said.
Hyper-converged infrastructure is an alternative to traditional SAN environments. HCI systems are sold as appliances, usually in three-node minimum configurations, that integrate compute, networking and storage. Those IT resources are managed through an onboard hypervisor. Interest in HCI spiked last year in relation to COVID-19, as organizations sought ways to enable employees to work remotely.
However, the inability of HCI to separately vary the ratio of compute and storage prompted some vendors to introduce a hybrid variant known as disaggregated HCI. Examples include NetApp HCI and HPE Nimble Storage dHCI.
NetApp HCI uses its SolidFire all-flash storage system attached to servers to provide a scale-out system. The Nimble dHCI runs VMware on Nimble Storage hybrid SANs, but Nimble doesn't run vSAN pooled storage. The NetApp and Nimble products allow customers to add compute or storage, although the architecture more closely resembles converged infrastructure than HCI.
Dave Raffo, a senior analyst at Evaluator Group, based in Boulder, Colo., said the enhanced VMware HCI Mesh is the vendor's move to address scalability concerns about HCI.
Dave RaffoSenior analyst, Evaluator Group
"VMware HCI Mesh enables independent scaling of compute and storage, which has been the big knock on HCI from the start. If someone only needed compute but had enough capacity, or needed more storage but had enough compute, they had to buy nodes that had both," Raffo said.
More vSAN use cases
Caswell said vSphere 7 enables customers to implement a brownfield HCI deployment and methodically migrate data from SAN environments. Another potential use case, Caswell said, is using vSAN as external storage for blade servers to reducing CPU licensing for core databases.
Along with HCI Mesh enhancements, VMware also doubled the scale-out density of a vSAN data store from 64 nodes to 128 nodes. Its vSAN Based Policy Management tool expanded to include compute clusters, enabling users to automate data management policies and apply the storage attributes to select VMs.
VMware also added features to support the rise of AI workloads at the edge. That includes the ability to resize persistent data volumes without disruption and expansion of stretch clusters to 20 nodes for primary and secondary data, up from 15 nodes previously.
This week's launch marks an extension of VMware's Nvidia partnership to build enterprise-ready AI infrastructure. It coincided with Nvidia's rollout this week of an AI management framework certified to to run exclusively on top of vSphere 7.