Deploying vSAN
vSAN allows for per-VM policy-driven storage provisioning. vSAN provides tight integration with vSphere, allowing VMware Cloud Providers offering a hybrid cloud service to easily utilize vSAN in the storage infrastructure.
The remainder of this section provides an architectural overview of a hybrid cloud model comprised of a VMware Cloud Provider data center providing dedicated hosted resources for their business customer, Rainpole, a hypothetical company used to illustrate potential use case.
Rainpole is preparing to stand up new infrastructure to be used to host a hybrid cloud environment. This particular VMware Cloud Provider is launching a new service based on vSAN, and Rainpole’s new infrastructure appears to be a good use case.
This infrastructure will host a new application, therefore, Rainpole does not have any historical data, so they are designing from scratch. Rainpole’s infrastructure requirements are as follows:
• 400 virtual machines
• Each VM has a 50-GB VMDK and 8 GB of virtual memory
• Tolerate 1 failure and support 1 snapshot
• 70% read, 30% slack space, and 20% future growth
Using this data along with the vSAN sizing tool, the service provider has calculated the infrastructure required to support this inaugural customer as the following:
• 20 nodes (2x Intel Xeon Ivy Bridge 8C E5-2640V3 2.6 G, 8x 16 GB 1866 MHz DDR3 SDRAM, 1x LSI 3008 based controller, 1x 10G support w/X540 chip with 2x RJ45)
• 80 HDDs (1 TB, 7.2K RPM, NL-SAS, 2.5-inch drives)
• 20 SSDs (200 GB, SATA)
The vSAN resource cluster is configured in the following manner:
• Each node is configured with 1 disk group (1x SSD, 4x HDD) for a total raw capacity of 4 TB and total cache capacity of 200 GB.
• 2 resource clusters of 10 nodes for a total raw capacity of 40 TB and total flash capacity of 2 TB per cluster.
• Max IOPS per node is 8K and 80K per cluster.
The infrastructure is divided into two resource clusters for two reasons. The first is that a smaller cluster size, somewhere between 10 to 15 nodes depending on use case, provides a higher level of availability. Having all of the nodes in a single cluster provides a larger fault domain and increases risk. The second reason is that it provides the ability to scale both up (within nodes) and out (adding nodes to the cluster). In this example, each cluster is spread over three fault domains for enhanced reliability. See the following figure for a diagram of the cluster layout.
Figure 5. Rainpole Cluster Layout
