HCI Benching the lab

HCI Benching the lab

Just a small tidbit. I have built the vSphere/vSAN part of the lab and am curious as to what the performance is, so I’ll use HCIBench for that: https://flings.vmware.com/hcibench.

Deploying it is pretty straightforward, it’s an OVA:

After that (and powering on), the configuration is done through the web-interface:

Previous times I ran HCIBench at customers, I used VDBench as the bench-marking tool. In this version FIO is available, so I’ll give that a go.

The “Clear Read/Write Cache Before Each Test Case” is to make sure that the vSAN cache is cleared, which would impact the test results if it was not.

After saving the config, we validate it, which will tell us if we can move ahead:

So next, we start the test:

This will deploy the worker virtual machines to the cluster:

(and yes, I changed the theme on my vCenter server, to match the HCIBench (I definitely like the dark theme :)).

During the disk preparation, the CPU usage in the cluster is rising (as is the fan-speed:

and power usage (from approx. 160 to well over 300 Watts) within the physical box). The performance metrics on the vSAN Backend are definitely showing activity:

some errors in vSAN:

and high latency on the physical layer:

After about an hour, I decided to break off the test and try again, with VDBench. For this to work, we have to download VDBench manually (from https://www.oracle.com/downloads/server-storage/vdbench-downloads.html) and insert it into HCIBench:

After this, save the config and validate it again and start the test. This time, I deselected the Easy Run, since it will randomize the storage and that was what was taking so long (and intensive). So finally, after the redeployment of the workers, we get some results… Unfortunately, since my pfSense stopped working during this test (it spiked on CPU usage and I tried to do a reboot but got stuck in a boot screen), I was not able to get the cool Grafana results, but the results from after the test showed me that my setup might be okay for a lab, but is definitely lacking performance for “the real world”:

No harm though, this was not unexpected at all.

Please follow and like us:

Leave a Reply

Your email address will not be published. Required fields are marked *