vSphere 6.7 Will Not Run In My Lab: A Parable

CPU Icon“Hey Bob, I tried installing vSphere 6.7 on my lab servers and it doesn’t work right. You tried using it yet? Been beating my head against a wall here.”

“Yeah, I really like it. A lot. Like, resisting the urge to be irresponsible and upgrade everything. What are your lab servers?”

I knew what he was going to say before he said it. “Dell PowerEdge R610s.” I was actually surprised it was that new, and rack-mountable.

“Yeah, you’re out of luck. CPUs before the E3/E5/E7 family didn’t have VT-x extensions in them to make virtualization easy so VMware had to do this thing called binary translation. vSphere 6.5 was the last release that they supported that on because, frankly, it’s slow and everything associated with that technique is getting really old.”

“What the hell? You’d think they’d tell people about that!”

“What, an obscure KB article with absolutely no practical information in it and a reference in the 6.5 release notes to said obscure KB article didn’t catch your eye?” I say, dripping with sarcasm. “I think there was a warning that flashed on the console of affected hardware when you booted, too, but to be honest I only know that because someone mentioned it, I’ve never seen it myself.”

“That’s total crap, like anybody looks at the console. So now what am I going to do? All my gear doesn’t work.”

“One might argue it works just fine. 6.5 will be supported until November of 2021, you could stay on that. You could run 6.7 nested inside 6.5. I know this is a terrifying thought but you could buy some new equipment, too, something that was on a HCL this decade. Given the current generations of CPUs you’d probably be able to cut your VMware licensing in half while doubling your performance. Stick it to the man, or something.”

“Ha! Somehow I doubt my six licenses would attract their attention. I think I’d need four anyhow for vSAN. Maybe I’ll try the nested thing. Thanks man.”

As a side note to my parable here, if you’re thinking about this and have some time before you have to refresh your hardware it’s worth waiting to see how all this Spectre/Meltdown stuff turns out. None of the junk the ferenghis at Intel are shipping today is secure, at any level, especially given the latest wave of vulnerability disclosures. AMD might also turn out to be a good play moving forward, too, if they’re not in exactly the same spot because they blindly copied everything from Intel. The SSD shortages are subsiding so you don’t have to plan 60 or 90 days out anymore. Time will tell, so take some time if you can.

Comments on this entry are closed.

  • Funny narrative Bob!
    All of my home lab hosts are incompatible.

  • The explanation doesn’t really add up….. Nehalem and Westmere-EP CPUs like the all the CPU Options for the R610s, have the Intel VT feature.

    Virtualization of 64-Bit OSes has not been supported for a _LONG_ time on non-VT-x supporting CPUs, and all Windows server OSes are 64-bits.

    For some reason…. ESXi 6.7 no longer works on a lot of CPUs, and there doesn’t seem to be any satisfactory explanation for why.

    • Agreed, 100%.

      Binary translation might not be the biggest factor in dropping support for large numbers of CPUs. VT-x has been around since Nehalem, yes, but VT-d (virtualizing the IOMMU) hasn’t. From what I can see that’s what got added universally in the E3/E5/E7 v1s and newer. VMware also alludes to it in the release notes for 6.5 when they say “vSphere 6.5 is the final release that supports Software-Based Memory Virtualization. Future vSphere releases will not include Software-Based Memory Virtualization.” They may be speaking of Extended Page Tables, though.

      The biggest problem here is that VMware hasn’t said anything about this, which sucks. I don’t know if it’s because it’s too technical for marketing, or if it’s the combo of being too technical and being negative news, but it’d be handy if they said something other than hiding behind the HCL. I get why reams of old server models aren’t on the HCL, but not being on the HCL doesn’t necessarily mean something doesn’t work, it just means it isn’t supported. In lieu of VMware actually explaining themselves the HCL was our best bet to derive where the line was drawn, and we can’t.

  • Great article as always, Bob.

    How sad is it that vSphere 6.7 doesn’t work on my three old-ish HP Proliant DL360 G7’s, but Windows Server 2016 and Hyper-V works just fine? Argh…