When selecting a big hard drive for your homelab's VMware ESXi VMFS datastore, consider the 512e over the 4Kn version

Posted by Paul Braren on Feb 17 2021 (updated on Feb 18 2021) in

support-policy-4k-sector-hard-drives — Image adapted from "Microsoft support policy for 4K sector hard drives in Windows"

In my day job in IT, the move to all flash has been pretty much complete for all near big business and enterprise sales for a good 2-3 years now.

The picture is different in the home lab. NASs are commonly still using 3.5" HDDs, and VMware hosts such as the Bundle have plenty of room for big HDDs formatted as VMFS datastores. In my home lab, my use of such HDDs has been largely relegated to archive and backup purposes, such as:

Windows Server network shares
NAKIVO backup repository
Veeam backup respository
So honestly, the performance hit didn't matter too much, but your results may vary widely.

I found three good articles that explain why and how we've moved past ancient 512 byte sector sizes and onward to 4K Native and 512e drives, aka AF (Advanced Format) 512-byte Emulation. Here they are.

Dell

512e and 4Kn Disk Formats - published in 2017, excerpt:

This paper provides the context for 512e and 4Kn disk format migration, as well as pointing out the long-term benefits to customers and potential pitfalls to avoid when moving from 512-byte to 4K sector formats.

VMware

FAQ: Support statement for 512e and 4K Native drives for VMware vSphere and vSAN (2091600) - published in 2018, excerpt:

vSphere/vSAN 6.0 and earlier versions have not been designed to use 4Kn/512e direct attached disk drives. 512e drives are supported only in version 6.5, 6.7, 7.x and later.
Performance Best Practices for VMware vSphere 6.7 - published in 2019, excerpt:

4K native (4Kn) and 512B emulation (512e) drives have some advantages and some limitations. In particular, such drives can perform poorly if your workloads donʹt issue mostly 4K-aligned I/Os. For more information on this subject, see VMware KB article 2091600...

When shopping for big HDDs, 512e or 4Kn?

So, why would you choose one over the other? Well, those articles above don't really get into that, basically contact the vendor of whatever software workloads you plan to have saving data on those drives. It's a home lab, who knows what I might want to repurpose the drive for.

About 2 years ago, I remember doing a quick test of my new Helium filled HGST HE10 10TB drive that arrived as the 4Kn version instead of the 512e of my existing 2 otherwise identical drives. In other words, I believe the Amazon listing may have been wrong, example screenshot below.

Alas, it was too late for me for a return. While I wasn't exactly thrilled with the results of this one synthetic benchmark, I moved on, and put it to work doing daily scheduled backups.

See also HGST HE10 product selector pdf, pictured below.

Conclusion

data-sheet-part-numbers-ultrastar-dc-hc510 — Click to view HGST PDF.

Learn from my mistake, check those labels carefully, as it can be quite difficult to be sure the vendor will sell you EXACTLY what was listed online. It's better to check the manufacturer of the drive to locate the exact part # to shop for.

You may also want to test those drives out before putting them to work, holding on to your precious digital life for you.

You may also want to avoid SMR drives if you care about performance, see also Lack of SMR disclosures in the HDD business.

So here I am again 2 years later, finally revisiting this 512e/4Kn topic with a quick lab test that I decided to record. Why now? I'm preparing for a rearrangement of most of my VMware local 3.5" drive based datastores which would make far more difficult to do this test ever again, as I move from one Xeon D cluster node to the other.

My enviroment was the popular Supermicro SuperServer Bundle, a versatile Xeon D based system running fully supported VMware vSphere / ESXi 7.0 Update 1d with a fresh Windows 10 VM 2H20 VM with the latest VMware Tools.

What I learned was that performance of ATTO Disk Benchmark differed dramatically during writes, especially at 4 KB length, see for yourself in the video.

VM performance on a VMware vSphere 7 VMFS 6 datastore on 512e and 4kn drives

I realize this is far from a perfect test, it's focused on a single user home lab use case. This has little to nothing to do with like a vSAN use case where you'd likely want to run something like HCI Bench instead, comparing multiple 512e and 4Kn drive groups.