Scalability Improvements for the ESA in vSAN 8 U2
vSAN has always had scalability at a core focus on goal. You can easily expand a cluster by adding nodes, and add more compute, storage and networking capacity to handle more workloads. It also has supported scaling vertically for some monster VM's needing 8 socket hosts.
Another one of the hallmarks of the vSAN ESA is its efficiency, not only in data storage, but data processing. With the reduced effort it takes to process I/O, we’ve been able to increase the supported number of VMs per host from 200 VMs up to 500 VMs per host. This massive increase in VM density will allow many of our customers to take advantage of the latest hardware to run more VMs on fewer hosts, and build all new high-density cluster designs when the hardware resources can accommodate for it. While this can apply to all workload types, note that VMware Horizon is not yet compatible with the Express Storage Architecture. Beyond VDI, a common usecase for ultra scaling is automated testing environments that run hundreds of permutations of low overhead VMs.
How did VMware engineering deliver this improvement?
VMware vSAN delivers I/O using far fewer CPU cycles, freeing up processing for more density.
ESA uses a different snapshot engine, which does not create a new object (and new components) for each snapshot created. Given snapshot are a integrated enviroment in VDI, VCD, and vRA clone heavy enviroments this helps reduce concerns of component exhaustion.
While this limitation provides much more flexibility in cluster designs, good design principles still apply. One should still work through several considerations, including, but not limited to:
The proper virtual hardware specifications for VMs given the needs of the organization.
The proper CPU, memory, and storage capacity to support the desired number of VMs per host.
How demanding are the given VMs (workloads, virtual hardware assigned, etc.)
How many VM instances does the customer want to withstand in a host outage in the event of a host failure.
Networking requirements to satisfy desired performance capabilities, host evacuation times, etc