Time to rework my VMware vSAN based Home Lab over the next weeks. Question: I still have an Intel Optane 900P PCIe 480 GB disk (from 2017) in each server as a vSAN caching disk. Would these days a Samsung 990 Pro M.2 disk a better choice from a performance perspective?
@Aschenbrenner the Samsung 990 does not have Power Loss Protection. The PM9A3 do.
@ErikBussink @Aschenbrenner I would probably continue with the Optane if it has enough space.
Sure, there is a huge theoretical advantage for the Samsung, but in real life I'm not sure you will notice. And the 990 is a consumer drive that might get slower as it fills up.
@ErikBussink @Aschenbrenner
I can mention that I had a Optane P5800X for tempdb, it still had a couple of ms latency.
I changed job, so unfortunately I can't play with it anymore. 😪😪😪
@Aschenbrenner @fb I’m using Intel P4801X 100GB for the caching tier. Good enough for my lab.
@ErikBussink @fb Does the PLP availability make a difference in vSAN? I think that someone told me back in 2017 that vSAN doesn‘t give you the best possible performance when it detects that the caching tier has no PLP support. Am I right on this one?
@Aschenbrenner @ErikBussink
Sorry, I have no real experience with vSAN. My gut feeling says Optane is best, but you could always buy the 990 PRO, and put it in the gaming rig if you are not happy with the server performance? 🙂
@ErikBussink I don‘t want to bother you, but do you know here something? Thanks 🙏
@Aschenbrenner No i don’t.
@Aschenbrenner I’m asking this further and will get back to you if i have an answer.
@ErikBussink I've just read now about the new vSAN ESA, and learned that there is no difference between cache and capacity disks anymore. So, the question is now if the PLP-enabled SSD still makes sense in a Home Lab scenario with vSAN ESA? In addition, I can also meet the 25 GBit network requirement :-)
@Aschenbrenner are you using vSAN for normal storage or are you using FileServices or Stretched policies ?
@ErikBussink I'm using vSAN for storing my VMs - nothing more.
@ErikBussink Would it then work in my scenario?
@ErikBussink Cool - thanks! My plan will be to use in each ESXi host (3x) a Dual M.2 PCIe SSD card with a 1 TB and a 2 TB Samsung M.2 NVMe (960 & 990).
@Aschenbrenner YMMW if you are going to mix different generations of SSD, your perf will be all over the place.
Remember that vSAN ESA requires a vSAN Readnode on the VMware HCL.
There is also a memory footprint that will be consumed for vSAN. hope you have enough RAM.
@ErikBussink Of course everything is not on the HCL and gives you warnings during the installation/configuration, but who cares about it in a Home Lab scenario?
@Aschenbrenner While not on an HCL, you do care to have a stable environment that is predictable and highly available.
A Home Lab can run important workloads for you, for training and for demos.
One point you mention, is the mixing of Samsung 960 and 990 Pro. There is a lot of perf difference between those SSDs. I would not do that in my lab.
@ErikBussink So you would go just with one version - like the 960? Thanks for the input!
@Aschenbrenner Yes stick with one type of SSD.
The 960 are the Pro version right ? Not the Evo ones.
@ErikBussink I will go now with the 990 Pro's (2 TB). In the first step I will *try* to mix them with the existing 960 Pro's (1 TB). If I have problems, I will swap them also to 990 Pro's (1 TB). Thanks for your help & time! :-)
@ErikBussink But from a performance perspective, the limiting factor is the PCIe Gen 3.0 interface of the HP DL180 G8 server.
@Aschenbrenner #vSAN does not recognizes if SSD have or have not PLP. That is the reason why the #VMware Compatibility Guide (VCG) sticks with devices only using PLP.
@ErikBussink My Optane 900P has also no Power Loss Protection. This was a little overkill for a Home Lab scenario ;-)

@Aschenbrenner @ErikBussink
I think I've seen a picture of your home lab on another site. I think it was "pretty serious". =)

But the Samsung PM9A3 that @ErikBussink mentioned earlier also looks very interesting (stable, PLP, reasonable price etc.)

@Aschenbrenner That would depend on the workload. A Samsung 990 PRO will do much better for sequential throughput, but that old Optane will have lower latency and better random perf at low queue depths