We tested the scalability of eight VMs across the four-node vSAN Ready Node cluster. We tested one VM, and with each iteration of the test, we added a VM in the following manner:
Figure 5. Test iteration layout
Each VM was configured with the following parameters:
Table 3. VM configuring parameters
Memory for VM
320 GB (operating system)
Memory for SQL Server
256 GB (SQL Server)
Filesystem on Red Hat Enterprise Linux
1 TB TPC-C
VMware PVSCSI controller
HammerDB is an open-source tool that is used for benchmarking and load testing different databases including SQL Server. We used the TPC-C benchmark, which is a widely accepted OLTP benchmark. A scale factor of 10,000 warehouses resulted in a SQL Server database approximately 1 TB in size.
The following table summarizes the test parameters:
Table 4. Testing parameters
Number of warehouses
Transactions per user
Keying and Thinking time
Enabled for 10% of all transactions
Keying and Thinking time was enabled for our testing. However, using the HammerDB tool, only 10 percent of the transactions were subjected to a total Keying and Thinking time of 1 second, rather than the default high values, ranging between 20 to 40 seconds. These values were chosen to simulate a high-stress workload.
The goal of the testing was to simulate high-stress OLTP SQL Server database loads that stressed the vCPU to within 50 to 60 percent utilization. This test was monitored and verified using dstat captures for the VMs. These VMs simulate the behavior of eight production SQL Server VMs in an enterprise.