Microsoft's licence riddles give Linux and pals a free ride to virtual domination Microsoft is caught in a monkey-trap, created by cloud computing and Free Software, coupled with short-term thinking and a dose of not-invented-here syndrome. You know how monkey-traps work? You make a small hole in a coconut shell, put some bait in it and tie it to something. The monkey comes along, reaches in for the bait …
As long as you set up the KVM vm correctly - i.e install in an LVM partition, set correct cache settings and use Virtio drivers it will outperform all other types of virtualisation.
You HAVE you use the Virtio Disk drivers (as your installing in an LVM partition) or the Windows installer will not see any drives btw.
(correct cache settings = native,none)
Try it, benchmark it and you will see
Not true. I have tried quite recently - Hyper-V and VMware both significantly outperform KVM, even when some effort has been made to performance tune KVM.
I suspect that this is because KVM has to run virtualisation as an added function under the Linux Kernel, whereas Hyper-V Server and VMware both use dedicated hypervisor layers.
This is not really news either - for instance both VMware and Hyper-V can reach over 1 million IOPS in a SINGLE VM - and did so well over a year ago. KVM as yet can't.
Did you house your Windows VM in an LVM partition, instead of a QCOW image?
Only when I do that coupled with the cache settings NONE/NATIVE does Windows IO speed outperform other types of virtualisation - i.e you NEED to install inside an LVM partition (and use Virtio drivers to get 'proper' speed)
Also your wrong about IOPS. KVM currently holds the IOP's world record...
http://forsetti.wordpress.com/2012/06/28/kvm-technology-review-and-roadmap-update/
'KVM has achieved World Record IOPS: 1,402,720 IOPS on a IBM x3850 X5 for 8KB request using 7 SCSI pass-through devices.. For 1 KB requests, can achieve 1.65M IOPS.'
"Also your wrong about IOPS. KVM currently holds the IOP's world record..."
That's a laughable test with *MULTIPLE* VMs - just piling it as high as they can - KVM can't manage 1,000,00 IOPS in a *SINGLE* VM - which tests the true scalability of the hypervisor...
The more recent tests rom IBM that do claim to exceed 1 million IOPS are NOT using a production hypervisor. Hyper-V and VMware both did this on production releases nearer 2 years ago. KVM is desperately throwing beta software and the latest hardware at the gap to try and catch up with what competitors did 2 years ago...Not to mention that this performance is only achieved in a crippled environment where most of the hypervisor features like storage migration don't work!
"That's a laughable test with *MULTIPLE* VMs - just piling it as high as they can - KVM can't manage 1,000,00 IOPS in a *SINGLE* VM - which tests the true scalability of the hypervisor..."
[Redacted], Redmondian hyper-shill. Who the [redacted] runs one VM per host? KVM is good enough for Google, which makes it better than anything the rest of the world has. Google's tech > the rest of the world.
Added bonus: they don't posses an Empire of Sadness.
Also regarding IOPS -
ftp://public.dhe.ibm.com/linux/pdfs/KVM_Virtualized_IO_Performance_Paper_v2.pdf
" It also details test results that demonstrate that a single KVM
guest can handle more than 1.2 million I/O operations per second (IOPS) at 8KB I/O request
size and more than 1.5 million IOPS at 4KB and smaller request sizes – the
highest storage I/O performance ever reported in a virtualized environment."
So your statement regarding VMWARE /HyperV IOPS speed being greater than KVM was just plain wrong....
I can only assume your benchmarking was also.
It will outperform *because* of the cache, at the risk of data loss in the event of a power failure. Writes are confirmed when they hit memory, not when they are safely stored on disk.
If you disable the cache (which you should do for safety) it's usually outperformed by others.
I'm not up on MS licensing, so can't really comment, but from a position as someone who works for one of the big three backup companies we see a high demand for Hyper-v client and development of that client. I was at a meeting the other day where one of the execs described Hyper-v as "coming through like a freight train."
It's also not the case that you can't easily chop Windows down to next to nothing. I run a small lab at home with VMware and Hyper-v hypervisors, the Hyper-v machine has a tiny footprint and is command line only, with practically nothing installed, I can take it all the way from there up to a full server with GUI, server services, applications , etc.
As someone who is up on Microsoft licensing I can only say that there are reasons why I refer to Microsoft's licensing department as the Empire of Sadness.
After war criminals and a few select violent sociopaths, those who people the Empire of Sadness are, in my opinion the worst individuals our species has to offer.
This was solved years ago. Buy a Windows DataCentre edition license for each ESX or Hyper-V platform. That then allows you to deploy unlimited Windows Server OS on top. Job done.
We saved a fortune on massive virtualisation estates by doing just this and far more importantly, simplified the whole licensing tracking & planning.
Hyper-V Server does NOT require a paid license. It is totally FREE with all features included.
Licensing client VMs is exactly the same under any Hypervisor - be it Linux, VMware or Hyper-V.
Therefore this article is a load of misleading FUD.
"Microsoft doesn't recommend the freeware Hyper-V Server for serious workloads"
Just utter rubbish. Hyper-V server is recommended for high end workloads as you don't have the overhead of Windows Server.
"Microsoft doesn't recommend the freeware Hyper-V Server for serious workloads"
Wrong - See - http://blogs.technet.com/b/keithmayer/archive/2012/09/07/getting-started-with-hyper-v-server-2012-hyperv-virtualization-itpro.aspx#.UqmvJaOnyUk
As an enterprise-grade, bare-metal hypervisor solution, Hyper-V Server 2012 offers the same level of scale, clustering, live migration and DR-replication capabilities as the Hyper-V role in Windows Server 2012, but at an unmatched feature/cost price point in the industry - an enterprise hypervisor feature set for FREE
Why would I use Hyper-V Server 2012 instead of Windows Server 2012?
In some scenarios, you may not need the additional capabilities of Windows Server 2012. For instance, you may be looking to setup dedicated hosts running only the Hyper-V role for consolidating large numbers of virtual machines on standalone or clustered hosts. In some cases, you may also not need the additional license grants provided by Windows Server 2012 - you may have already purchased your Windows Server licenses, or you may be setting up a client virtualization/VDI solution leveraging Hyper-V where Windows Server OS licenses are not needed for each VM. In these scenarios, Hyper-V Server 2012 makes perfect sense to deploy and integrate into your enterprise network.
No problem - but they've changed their tune, then. My source was the Microsoft speakers at _Microsoft's launch event for WS2012_ which I covered here:
http://www.theregister.co.uk/2012/11/01/windows_server_2012_technical_launch_day_overview/
... and here:
http://www.theregister.co.uk/2012/11/02/windows_server_2012_what_is_new/
Datecenter edition licenses are a good option if your budget can cover them. It would make sense in a green field or an large enterprise. I have always worked in mid sized companies where projects are deploying a handful to a dozen servers at a time. The way the budgets are done just won't allow the huge expense of Datacenter edition for an entire cluster. When I first deployed ESX for a large(ish) VMware cluster to replace an all physical data centre we still re-used the existing licenses as we virtualised. In the long run you end up paying over the odds as Standard (or in the past Enterprise) versions are bought piecemeal, which MS must be well aware of and happy about!
There's more to licensing re: vritualisation than the OS shenanigans.
Yes. It's VDI licensing. I believe I was very clear about that. It doesn't change the fact that Microsoft licensing is a bithc, nor that every single person who works for The Empire of Sadness should [redacted] into [redacted] and then [redacted] [redacted] [redacted].
And the horse they rode in on!
Interesting article and I can most certainly relate to all this. I'm currently using 2 in-house Windows 2003R2 servers and one 2008R2 for an ASP project and we ended up with the same conclusion. There are so many loopholes and uncertainties; it even seems as if the Microsoft sales agents also have no clue other than to advice you to go for the most lucrative licensing scheme (lucrative for them of course).
It's depressing, especially if you're actually interested in the environment(s) and the technique.
But the bottom line is quite simple really: Microsoft doesn't know how to appeal to the masses. You would think that after so many years with several competitors around (even those which cannot be bought) they'd smarten up, but no. There have been some very good moves (some commercials are very slick, some of the new features are very well designed too, etc.) but those are merely bits and pieces. In the overall it's one unappealing mess.
Think about it: this is yet another situation in which their TechNet environment might have been able to help them (to a certain degree). For example; I hope we all know that TechNet is/was all about systems administration; you could lay your hands on pretty much anything which Microsoft provided and use it in test labs and such.
They could have raised the bar a bit I think. Provide a "TechNet virtualisation" license which allows subscribers to use an x amount of specific client / server products within a (fully licensed) Hyper-V environment. It may certainly appeal, it makes things easier and the most important thing: it would generate more steady revenue for Microsoft.
Bet they never thought of that, not to mention having little options left considering how they're going to whack TechNet instead of extending on it.
They need to make it easier for the masses, not harder.
"They could have raised the bar a bit I think. Provide a "TechNet virtualisation" license which allows subscribers to use an x amount of specific client / server products within a (fully licensed) Hyper-V environment."
TechNet (and MSDN) already does allow that.
Technet's dead, Jim. And MSDN is beyond the means of your average non-enterprise sysadmin and many SMBs.
Nope, IBM VM is something quite different again - this is using one OS kernel to multitask multiple instances of another, different OS kernel, one per user. This is directly homologous to, say, running multiple Windows Server instances on VMware, or running multiple Linux instances on top of Hyper-V Server.
Again, do buy the ebook. :-) I'm not on royalties here! It's just the best intro to the basic underpinnings that I could do.
I don't need to read the ebook, I lived it. It was a really long time ago when I was running VM (and VS1) on an early 4341 but I think I remember how it worked.
CP by itself was not useful - it only visualized the hardware.
You needed a real OS to perform any work.
The interesting thing was CP was so good that VS1 would run better on it then on the native hardware (CP handled memory/paging better then VS1 could).
No, timesharing is something quite different. May I suggest my Reg ebook?
http://www.amazon.co.uk/Brief-History-Virtualisation-Liam-Proven-ebook/dp/B006MSECAS
The source articles are still on the Reg if you search for them.
Timesharing means simultaneously multitasking multiple interactive user sessions; it's something totally different.
It's entirely fair. Unlike Windows 8, which was a counter-rotating clusterfuck of pain, shame and psychological abuse, Server 2012 is a truly excellent operating system that was significantly improved by Server 2012 R2.
Whereas the people responsible for the major design decisions of Windows 8 deserve to be [redacted] with [redacted] and [goats, again, really?] and then [nobody deserves exposure to the Empire of Sadness. Not even the endpoint team], the Server geeks did a fucking fantastic job and delivered a great operating system worth the full price upgrade.
It was not, however, worth the $2000 price hike on top of the full price, nor the gutting of SPLA of the annihilation of partner rebates.
But hey, my CentOS installations have gone up 2400% in the past 4 months. The price of rice is just too much for China...
Okay, *YOU* are a Microsoft shill. Flat out. The bald-faced assumption that everyone pays a subscription fee and presenting this not only as a fait accompli but "normal" (without any comment or provision for the other 80% of businesses in the G8 nations) shows you attempting to push Micfosoft's preferred vision of the world in defiance of what the majority of the world actually looks like.
In fact, you're probably an operative of The Empire of Sadness. I hope you [redacted] and [redacted] you [redacted].
"POSE" could stand for...
pAIN oN sTART eVERYWHERE // Pain On Start Everywhere.
How about a license called "PLIGHT" -- Possible Litigation Is Going High-Time...?
On the serous, umm, SERiOUS side (seriously!), how long before virtual OSs will come to "green, disposable" (flushable/dissolvable) micro appliances so that users can either via wire or bluetooth or other air-delivered protocol surf with less need for worry about device (laptop, desktop, phone, household hardware)? Such virtual machines could be interfaces in the physical, too, either atttached to or sitting between individual devices, akin to a supercomputer-based anti-virus, anti-malware, anti-anything-you-despise-trying-to-get-on-your-devices firewall.
That is true of /all/ forms of virtualisation, though. If you're running dozens of Windows or Linux or whatever instances on top of a single copy of Windows Server with Hyper-V and that host copy of Windows goes down, all your instances are gone, instant toast.
OS virtualisation makes no difference to this at all.
Of course, this is what clustering and failover are for, but then again, if Windows integrated some form of containers, there is no reason at all that they could not fail over to containers on other hosts.
While it's true that in the case of hypervisors like ESX or Hyper-V the guest OS's are at the mercy of the hypervisor, it's still a better solution.
OS partitioning/virtualization techniques like Solaris Containers will never be popular (IMHO) because the
management is too difficult. Sharing the OS and it's files means that every container must be taken down and patched - at the same time! This is practically an impossibility unless all your containers have the same management cross-section. This may be the case for some specific applications (VDI perhaps?), but makes creating a general purpose virtualization platform difficult!
I realize some containers have tools to assist with migrating the containers - sometimes live, but mostly off - to another physical machine/OS instance. Analogous to VMware's "maintenance mode", but probably a lot more manual.
Plus, as has been mentioned, the hypervisor can be as small amount of code as possible thereby reducing the bug and security fronts.
Personally believe that containers are dead. Long live hypervisors!
;-)
> No mention is made in this article that a single OS instance with "containers" running the apps, is a serious single point of failure.
Similar to a h/w failure of the host or a VM hypervisor exposing a previously unknown but deadly bug. You've got to have some sort of backup/disaster recovery in place.
The article fails to mention one important element: management. VMware's Virtual Center Server is a relatively cheap investment, with most of the cost being in the hypervisor licenses, so any serious business will buy Virtual Center. With Microsoft, by contrast, serious management is done through System Center 2012 Virtual Machine Manager. Anyone who has ever tried to license SCOM knows that SCOM licensing makes Windows licensing look sane and approachable. Nor is it cheap. In theory, Hyper-V is free, but if you want the enterprise-level features offered by VMware, you'll pay heavily for it.
Furthermore, if you want service and support, you will then pay even more for a Microsoft (or, more likely, a reseller) support contract and then pay even more for Software Assurance. I have not crunched the numbers, but it's simply unbelievable that Hyper-V offers a reasonable TCO on a feature-by-feature comparison.
This is true, yes, but I did specifically say:
> Then you need to licence the [...] layered products on top, such as Exchange or SQL Server.
> Of course, various bundles and deals apply to all this.
Those layered products include the high-end management tools.
It's a complicated, hairy mess, as many Microsoft resellers said to me in my background research.
"Anyone who has ever tried to license SCOM knows that SCOM licensing makes Windows licensing look sane and approachable. Nor is it cheap. In theory, Hyper-V is free, but if you want the enterprise-level features offered by VMware, you'll pay heavily for it."
Cost of Hyper-V + FULL System Centre Suite is MUCH cheaper than even base vSphere + Virtual Centre costs...And Hyper-V includes features for free like replication / site recovery that are expensive addons with VMWare...
Umm...
Sources are current retail prices as per VMWare and MS today, using today's exchange rates.
vSphere + Ops Mgr Standard is approx. £1320 for a licence and about £990 for 3 years of support. So around £2300 per CPU all in.
2012 R2 DataCenter is around £4500 and SystemCentre DataCenter is approx £2220, both prices including SA and are per 2 CPU's. That's around £3360 per CPU all in.
What the vSphere price doesn't take into account is the virtual machines you're running on that environment. Anything more than 3 Windows Standard (without SA) instances running and you're better off taking the 2012 R2 DataCenter licenses. If you're anyone who wants to achieve high levels of Windows virtualisation then you'll actually be better off buying R2 DataCenter in addition to your vSphere licenses.
Is MS licensing complex? In places yes. Is this one of them? No.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
