Thank you to the 424 (!) people who answered the survey that I started late on Friday afternoon and finished today (Tuesday morning). I asked one question:
What kind of UI installation do you use on Hyper-V hosts?
- The FREE Hyper-V Server 2012 R2
- Full UI
- MinShell
- Core
Before I get to the results …
The Survey
Me and some other MVPs used to do a much bigger annual survey. The work required by us was massive, and the amount of questions put people off. I kept this very simple. There were no “why’s” or further breakdowns of information. This lead to a bigger sample size.
The Sample
We got a pretty big sample size from all around the world, with results from the EU, USA and Canada, eastern Europe, Asia, Africa, the south Pacific, and south America. That’s amazing! Thank you to everyone who helped spread the word. We got a great sample in a very short period of time.
However (there’s always one of these with surveys!), I recognize that the sample is skewed. Anyone, like you, who reads a blog like this, follows influencers on social media, or regularly attends something like a TechNet/Ignite/community IT pro events is not a regular IT pro. You are more educated and are not 100% representative of the wider audience. I suspect that more of you are using non-Full UI options (Hyper-V Server, MinShell or Core) than in the wider market.
Also, some of you who answered this question are consultants or have more complex deployments with a mixture of installations. I asked you to submit your most common answer. So a consultant that selects X might have 15 customers with X, 5 with Y and 2 with Z.
The Results
So, here are the results:
70% of the overall sample chose the full UI for the management OS of their Hyper-V hosts. If we discount the choice of Hyper-V Server (they went that way for specific economic reasons and had no choice of UI) then the result changes.
Of those who had a choice of UI when deploying their hosts, 79% went with the Full UI, 5.5% went with MinShell, and 15% went with Server Core. These numbers aren’t much different to what we saw with W2008 R2, with the addition of MinShell taking share from Server Core. Despite everything Microsoft says, customers have chosen easier management and troubleshooting by leaving the UI on their hosts.
Is there a specific country bias? The biggest response came from the USA (111):
- Core: 19.79%
- MinShell: 4.17%
- Full UI: 76.04%
In the USA, we find more people than average (but still a small minority) using Core and MinShell. Next I compared this to Great Britain, Germany, Austria, Ireland, The Netherlands, Sweden, Belgium, Denmark, Norway, Slovenia, France and Poland (not an entire European sample but a pretty large one from the top 20 responding countries, coming in at a total of 196 responses):
- Core: 13.78%
- MinShell: 4.08%
- Full UI: 82.14%
It is very clear. The market has spoken and the market has said:
- We like that we have the option to deploy Core or MinShell
- But most of us want a Full UI
Those of you who selected Hyper-V Server did not waste your time. There are very specific and useful scenarios for this freely licensed product. And Microsoft loves to hear that their work in maintaining this SKU has a value in the market. To be honest, I expect this number (10.59%) to gradually grow over time as those without Software Assurance choose to opt into new Hyper-V features without upgrading their guest OS licensing.
My Opinion
I have had one opinion on this matter since I first tried a Core install for Hyper-V during the beta of Windows Server 2008. I would only ever deploy a Full UI. If (and it’s a huge IIF), I managed a HUGE cloud with HA infrastructure then I would deploy Nano Server on vNext. But in every other scenario, I would always choose a Full UI.
The arguments for Core are:
- Smaller installation: Who cares if it’s 6GB or 16 GB? I can’t buy SD cards that small anymore, let alone hard disks!!!
- Smaller attack footprint: You deserve all the bad that can happen if you read email or browse from your hosts.
- Fewer patches: Only people who don’t work in the real world count patches. We in the real world count reboots, and there are no reductions. To be honest, this is irrelevant with Cluster Aware Updating (CAU).
- More CPU: I’ve yet to see a host in person where CPU is over 33% average utilisation.
- Less RAM: A few MB savings on a host with at least 64 GB (rare I see these anymore) isn’t going to be much benefit.
- You should use PowerShell: Try using 3rd party management or troubleshooting isolated hosts with PowerShell. Even Microsoft support cannot do this.
- Use System Center: Oh, child! You don’t get out much.
- It stops admins from doing X: You’ve got other problems that need to be solved.
- You can add the UI back: This person has not patched a Core install over several months and actually tried to re-add the UI – it is not reliable.
In my experience, and that of most people. servers are not cattle; they are not pets either; no – they are sacred cows (thank you for finding a good ending to that phrase, Didier). We cannot afford to just rebuild servers when things go wrong. They do need to be rescued and trouble needs to be fixed. Right now, the vast majority of problems I hear about are network card driver and firmware related. Try solving those with PowerShell or remote management. You need to be on the machine and solving these issues and you need a full UI. The unreliable HCL for Windows Server has lead to awful customer experiences on Broadcom (VMQ enabled and faulty) and Emulex NICs (taking nearly 12 months to acknowledge the VMQ issue on FCoE NICs).
Owning a host is like owning a car. Those who live in the mainstream have a better experience. Things work better. Those who try to find cheaper alternatives, dare to be different, find other sources … they’re the ones who call for roadside assistance more. I see this even in the Hyper-V MVP community … those who dare to be on the ragged edge of everything are the ones having all the issues. Those who stay a little more mainstream, even with the latest tech, are the ones who have a reliable infrastructure and can spend more time focusing on getting more value out of their systems.
Another survey will be coming soon. Please feel free to comment your opinions on the above and what you might like to see in a survey. Remember, surveys need closed answers with few options. Open questions are 100% useless in a survey.
What about Application Servers?
That’s the subject of my next survey.
Using This Data
Please feel free to use the results of the survey if:
- You link back to this post
- You may use 1 small quote from this post
Pitty that I missed the survey, but we figured out the same along the way. Starting off with core only to find out that especially NIC issues were a disaster to figure out due to the lack of access to driver settings pages. So yeah, when we moved to 2012R2, everything was build with Full GUI. Why make administrating life difficult when you need your valued time translating new functional requirements into technical designs?
Honestly, now that remote administration is activated by default (ok, run winrm was not so difficult before) this is much easier to figure out what’s going on a host. Most of time perfmon, event viewer and cluster manager are the only Common tools needed. I do agree that it more difficult to use command line to perform a network dump than using wireshark or message analyzer oob, but that’s anecdotic !
Maybe that typical to Swiss market, but here most of my customers have bought SCCM when it was time to migrate Windows 7/8, so using VMM is not an issue.
Personally, I prefer core server and I’m pretty excited to see what’s coming with nano server ! (and if it’s compatible with cluster functional levels from 2012r2, that should surprise me !)
The stats above show you are not alone but not in the majority.
Aidan,
I cannot tell you how much I love you right now. You are spot on. Stay mainstream and stop doing stupid crap and you’ll have less chances of being best friends with 2am.
JamesNT
Hey Aidan,
These results do not surprise me, but I am at a tipping point at this stage in my thinking of management. This is a dangerous place to be and I have been burnt in the past – ServerCore on 2008R2 was that much harder to fix stuff – nvpbind.exe anyone?
For me though (disclaimer now in the DevOps space) I’m very attracted to infrastructure as code – I feel that this does solve a number of things, but I’m unsure if this is fully achievable on the windows server stack.
It feels wrong though to base the management decisions on you system to how well drivers work, or in this case don’t work, but understand the practicalities here. My only concern is if we always do what we’ve always done, how can we expect things to change?
Are vendors still gonna flick us off when everyone knows the driver is bad?
And what happens on the other side of the fence? Do such driver issues occur in the *nix world? If so, how do they approach the issue? Can this approach be used for us in the windows world?
Would having a diff of the infrastructure (if IaC is used) actually be helpful in determining the issue? ( i.e. this config works, this does not, show me the diff.)
Reuben
I do not understand the argument that a UI prevents you from moving forward. You can easily do devops the same way if there’s a UI. You have PowerShell still. You just don’t have the risks with a UI.