Intel Optane SSD P4800X 750GB sample testing with VMware vSAN OSA

in a previous post, I talked about the Intel Optane SSD P4800X 750GB drives I received from Intel via the VMware vExpert program and testing them with VMware Horizon. In this post, I will be testing them with VMware vSAN in a 4-node cluster.

http://10.0.1.160/intel-optane-ssd-p4800x-750gb-sample-testing-with-vmware-horizon/

The hardware setup

The card will be set up in four Intel Gen 11 Extreme NUC. These hosts are identical in configuration.

• Intel NUC 11 Extreme Kit NUC11BTMi7
• G.Skill Ripjaws 4 F4-3200C22D-64GRS
• Lexar NM620 512GB (Boot)
• Lexar NM620 2TB (Capacity)
• Intel Optane SSD P4800X 750GB (Cache)
• 10Gtek® 10GbE PCIE Network Card for Intel X520-DA1 (Connected to Ubiquiti
  USW-Aggregation)
• 10Gtek® for Intel SFP+ Cable XDACBL1M
• Startech US1GC30B USB-C Gigabit nic

vSAN OSA setup

I have created a separate Distributed switch that has the vSAN and vMotion Networks. Then on this switch, I added the hosts with the 10Gbit card. The disk group will be consisting out of the Intel Optane SSD P4800X 750GB as Cache and a Lexar NM620 2TB NVME for Capacity.

In a later blog, I will also test vSAN ESA but for that, I still need to upgrade the Optane SSD P4800X firmware to the HCL version.

vSAN testing

This will be done using HCIBench.

HCIBench stands for “Hyper-converged Infrastructure Benchmark”. It’s essentially an automation wrapper around the popular and proven open-source benchmark tools: Vdbench and Fio that make it easier to automate testing across a HCI cluster. HCIBench aims to simplify and accelerate customer POC performance testing in a consistent and controlled way. The tool fully automates the end-to-end process of deploying test VMs, coordinating workload runs, aggregating test results, performance analysis, and collecting necessary data for troubleshooting purposes.

HCIBench is not only a benchmark tool designed for vSAN, but also could be used to evaluate the performance of all kinds of Hyper-Converged Infrastructure Storage in vSphere environment.

Im not going to explain the setup, this can be done via the manual of HCIBench. Basically, it’s deploying and configuration of the OVA and starting it up 🙂 Do some simple config and start the test. It automatically deploys 8 fio testing VMs.

Once it deployed all VM the testing starts and you can watch this live on a Grafana dashboard!

The HCIBench Results

The first test I have done is with the Workload fio-8vmdk-100ws-4k-70rdpct-100randompct-4threads-50compress-50dedupe.

PDF results: fio-8vmdk-100ws-4k-70rdpct-100randompct-4threads-50compress-50dedupe-1690999525-report

Datastore     = vsanDatastore
=============================
JOB_NAME	= job0
VMs		= 8
IOPS		= 95541.22 IO/S
THROUGHPUT	= 373.00 MB/s
R_LATENCY	= 3.47 ms
W_LATENCY	= 0.87 ms
95%tile_R_LAT	= 10.00 ms
95%tile_W_LAT	= 1.00 ms
=============================
Resource Usage:
+----------------------+------------+------------------+------------+
|                       Resource Utilization                        |
+----------------------+------------+------------------+------------+
| Management           | cpu.usage% | cpu.utilization% | mem.usage% |
+----------------------+------------+------------------+------------+
| vdr-esxmgt01.vdr.one | 56.57      | 29.47            | 89.97      |
| vdr-esxmgt02.vdr.one | 58.02      | 29.7             | 88.2       |
| vdr-esxmgt03.vdr.one | 65.75      | 35.42            | 93.12      |
| vdr-esxmgt04.vdr.one | 57.57      | 28.73            | 89.8       |
+----------------------+------------+------------------+------------+
| Average              | 59.48      | 30.83            | 90.27      |
+----------------------+------------+------------------+------------+


cpu.usage - Provides statistics for logical CPUs. This is based on CPU Hyperthreading.
cpu.utilization - Provides statistics for physical CPUs.

The second test is with the workload fio-8vmdk-100ws-256k-0rdpct-0randompct-1threads-50compress-50dedupe.

PDF results: fio-8vmdk-100ws-256k-0rdpct-0randompct-1threads-50compress-50dedupe-1691011148-report

Datastore     = vsanDatastore
=============================
JOB_NAME	= job0
VMs		= 8
IOPS		= 7826.22 IO/S
THROUGHPUT	= 1956.00 MB/s
R_LATENCY	= 0.00 ms
W_LATENCY	= 9.21 ms
95%tile_R_LAT	= 0.00 ms
95%tile_W_LAT	= 52.00 ms
=============================
Resource Usage:
+----------------------+------------+------------------+------------+
|                       Resource Utilization                        |
+----------------------+------------+------------------+------------+
| Management           | cpu.usage% | cpu.utilization% | mem.usage% |
+----------------------+------------+------------------+------------+
| vdr-esxmgt01.vdr.one | 47.55      | 23.05            | 92.11      |
| vdr-esxmgt02.vdr.one | 48.92      | 23.56            | 88.39      |
| vdr-esxmgt03.vdr.one | 48.31      | 23.51            | 93.2       |
| vdr-esxmgt04.vdr.one | 26.44      | 12.49            | 90.4       |
+----------------------+------------+------------------+------------+
| Average              | 42.81      | 20.65            | 91.03      |
+----------------------+------------+------------------+------------+


cpu.usage - Provides statistics for logical CPUs. This is based on CPU Hyperthreading.
cpu.utilization - Provides statistics for physical CPUs.

Conclusion

I have no other results to compare with this lab environment 🙂 unfortunately, I have no other SSD to test as a cache drive to see what the real difference is between Intel Optane and other drives.
As in the other testing with Horizon, I tested within a VM that is running on the vSAN datastore with Atto. Below is the result. So we can compare a VM on the Intel Optane SSD P4800X 750GB drive vs vSAN powered by Optane. See below.

Atto on vSAN

Atto on Optane

The vSAN config is stable and really performing well. With atto, the performance on vSAN is a lot quicker. However, the NVME capacity drives are also speedy. Be your own judge but my lab is loving the Intel Optane drives 🙂 Thanks again to Intel and the VMware vExpert community for making this possible.