In my previous post, we detailed the deployment and configuration of the HCI Bench to test our VxRail Cluster performance.
This post I wanted to go a little deeper into the “VDBbench Generate Page”. Some useful reference guides for this can be found here.
- Number of Disk to Test: – How many of the VMDKs / Disks from each Worker VM would you like to use during the test. Remember you set the number of disk per worker in the main UI
- Working Set Percentage: Working set = Number of Worker VMs * Number of Data Disks Per * Size of Data Disk. I think of this percentage as a data change rate in my Working set.
- Number of Thread Per Disk: Specifies the maximum number of concurrent I/O per Disk
- Block Size: – vSan has a native 4k block size. If you know what is your average block size is based on some performance analysis that you have completed you can change this value. One of the tools I use for performance analysis is Live Optics.
- Read Percentage: Specifies the read percentage of the workload again live optics can help here if you have real data
- Random Percentage: Specifies the random percentage of the workload
- I/O Rate: With this option, you can set a fixed I/O Rate. One of the reasons to do this is for testing mixed workload environments. IE How would my Horizon VDI World load run if I was already generating 100k IOPS from my standard VI world load.
- Test Time: How long the test will run for in seconds. This value will be overwritten by the time in the main UI if they are not the same.
- Warm Up Time: HCI bench will always ignore the first interval of results from the total. If you want to add additional warm-up time to allow the cache drives to fill up sufficiently you can add additional warm-up time here. This time will be appended to test time. Workloads will take a while for the cache to warm up before achieving steady performance
- Reporting Level: This parameter specifies the duration in seconds in each reporting interval. If you were running a particularly long Test over a number of days the amount of data that would be collected would be significant. by sampling the data at regular intervals it makes the data size more manageable.
One Last link I wanted to provide here was to a vSan benchmarking presentation that I saw earlier this year in VMworld that was very interested and insightful that was presented by Paudi O’Riordan and Cormac Hogan