VirtualVCP IT Virtualization

Now, if you do end up in a situation where you need to purchase additional licenses, the only thing that can think of to cushion the blow a little would be to double the amount of physical processor packages in each host, if you have the empty sockets available to do so.

At first it sounds like a good idea, but then you'll realize you've just moved the licensing cost to other products: Microsoft Windows Server Datacenter or RHEL licenses and any other 3rd party management tools (Quest/Veeam/vKernel/etc) usually licensed per-socket.

Is there a way to show the red line in the last graph to avoid confusion? I realize it's identical to the green line.

I don't think many customers are ordering servers (or blades) with empty sockets. That seems like an old school thing to do, right?

Also, the B230 UCS blade is a popular 2 socket blade, and can house 256 GB of memory with 8 GB DIMMs (up to 512 GB of memory if using 16 GB DIMMs). I don't think I'd buy one with only 128 GB of memory (or anything less).

Thanks, interesting comment.

Regarding the UCS blades and high volumes of physical memory that can be installed, I believe this is exactly the kind of thing VMware was thinking of when they came up with the licensing model.

In the case were you are going to install 192GB, 256GB and 512GB RAM in each balde and only have two physical CPU packages, VMware will be laughing all the way to the bank.

If you have 256 GB of RAM per blade and you wanted to allocate all of that memory to VMs, you will be required to purchase 6 vSphere 5 Enterprise Plus licenses where as in the past with vSphere 4, the license count would have been 2.

I can't believe that VMware actually went for low vRAM allocation figures such as 24, 32 and 48GB per license! Insane!

People can forget the hype about a VM with 1TB allocated memory, as it would be much more cost effective to run that server as a physical host. Allocating 1TB vRAM to a VM will require you to have 22 Enterprise Plus Licenses, and at a cost of almost $78,000.00 I can't see anyone being stupid enough to do that! VMware has shot itself in the foot here!

You're sort of ignoring the distinction between Allocated memory (which VMWare charges for) and Utilized memory.

It's not bad practice to _allocate_ 100%, or even more, of your physical memory to VMs. Between transparent page sharing, the memory balloon function, compression, etc., your Utilization might only be 65%.

So I can over-allocate memory, while actually underutilizing it. But Vmware is now going to charge me for the former. THAT is the problem.

I don't think it's fair to say "you're sort of ignoring the distinction between allocated memory and utilized memory"

Yes in hind sight, I should have renamed all the fields to Allocated memory and not Utilized memory, but this post is all about vSphere licensing and licensing doesn't care about utilized memory, but it does care about allocated memory.

And yes, in terms of HA and N+1, I do believe it's very bad practice to allocate 100% of physical memory in a cluster or host to VMs, even if it's not actually actively being used by the VMs. Then again, I probably don't deserve a VCAP-DCD badge for trying to keep to guidelines?? No?

Anyway, thanks for the comment. I should really go back and change my field to Allocated Memory as I can see how my post could be confusing.

Hi,
This is not commenting on the license, but on the RAM.

I agree with SNB that we can have vRAM = pRAM and we will be fine. A few reasons:
1. Configured RAM is not normally the peak RAM. If the peak RAM is 13 GB, the app team might ask for 16 GB. Easier to manage too, as 16 is a "familiar" number. And it feels saver to have buffer too :-)
2. Usage counter in vCenter is rather misleading. It is showing memory.consume counter, not memory.active. So it includes memory that _was_ active, but no longer. This is the reason why vCenter Ops will show RAM is fine while vCenter will complain RAM is high. Kit Colbert has good info at VMworld preso RAM.
3. If we have 10 VM in host, normally they don't peak at the same time.
4. TPS, Ballooning, Compression, etc.

So the above 4 reasons are enough to cover for HA. Naturally, I'm not assuming a 2-node cluster here :-)

Hope that helps
e1

Isn't the optimum memory configuration for a modern system going to be either 96 or 192 GB so you can maintain 1333 Mhz bus speed??? What do the calculations look like for those memory sizes?

Rynardt, good numbers as far as they go and as far as I can tell without going through your math.

However, we really need to take real-world scenarios into account. For instance, here you aren't including extra room for HA. So if your cluster is 4 + 1 spare, then you can reduce all these numbers by 20%, since you now have an additional host in your vRAM pool.

Thanks for the great calculations, Rynardt. Would you consider publishing your spreedsheet so we can plug in our own calculations.
We're deploying what we're calling a "high performance" cluster for those VMs that need greater vCPU and huge amounts of RAM and it would be interesting to see what 144 or 196GB RAM conparisons look like.

128GB RAM would pretty much the minimum when setting up hosts now.

I have customers who have 200GB+ memory in each host, just because their VMs need massive amounts of memory. Sure, those hosts may have 4 CPUs in them, but just using the Enterprise license (They don't need the extra features E+ grants them) they can utilize half of that memory. How can I explain (and justify) to them that VMware suddenly wants to double their license cost without giving them anything extra?

Also, I'd like to see how VMware handles those customers with the automatic SnS upgrade. If they have an environment that is compliant with the VS4 licensing, will they upgrade those customers so that they are covered even with VS5?

I agree with SNB, VMware charges for the allocation but not utilization, and in some cases you want to allocate more than you have (over commitment). You also state that is is a bad design and that HA gives you some overhead, but what if the customers isn't using HA (Say, it's an test environment with no uptime requirements) or that they want to test how their products (if they develop their own products) behave when memory is congested?

Just found out from a customer that they need to buy an extra 20(!!) E+ licenses in order to cover their entire environment.

They need 44 E+ licenses, that is an increase by almost 100% in order to cover their environment. They are now calling to a meeting where they will most likely replace VMware with Hyper-V, Xen or KVM.

Shame on you, VMware!

Don't jump into conclusions without doing proper math.

My examples cannot be applied to all environments as each and every environment will have different vRAM allocations and different actual active vRAM utilization.

Make sure you aren't over allocating and allocating more vRAM to VMs than what is actually required by the application within the guest OS.

Those numbers were straight from the customer itself. They allocate 100% of the physical memory

They have just purchased the following machines:

4 socket with 512GB ram.

VS4 = 4 licenses per machine
VS5 = 11 licenses per machine

Also, here are some real world examples from their environment:

vRam Details

Pooled vRam Capacity: 1152 GB
Number of VMs Powered On: 236
Current vRam usage: 1726 GB
Percent vRam usage: 149%

Number of VMs if all were Powered on: 288
vRam usage if all VMs were Powered on: 2005 GB
Percent vRam usage if all VMs were Powered on: 174%

As you can see, VMware will make big bucks on that customer, should they decide to upgrade.

Yeah, you see. This is the problem right here VMware. So much for VMware's real world examples that states we run 5.7 VMs per physical CPU and an average of 3GB per VM.

If you are going to allocate 1726GB to VMs and only have a pool of 1152GB, you will have to get 36 enterprise plus licenses. It's not just 11 per host, but 149% of 512MB per host could see you having to license at 16 enterprise plus licenses per host.

On this license model, the CPU socket count is irrelevant!

Hi Rynardt,

I've used your great post as inspiration for a calculation sheet for vSphere 4 vs 5.

See: http://www.azlan.cc/1/post/2011/07/vsphere-4-vs-5-dynamic-license-calc ulator.html

I've mentioned your blog as major inspiration.

grtz
Rob

Great article, your examples and calculation seem spot on to me and have to say, it re-inforces my view that VMWare have made a huge cock up.

One of the worst things about all of this is VMWare's spin that this is for the benefit of the customers. Utter rubbish. It's about generating extra revenue, especially as processors are going to come with more and more cores per socket.

It is a tax on vRAM, and taxes are used for two things, either changing behaviour or raising revenue. Nothing else.

From a big picture standpoint, this new pricing model is VMware saying Hyper-V and XenServer aren't competitive in the large enterprise. They're probably right and after six months of customer grumbling this price model will help support their future stock price.

Datto

Great post indeed.
However I personally would redo all calculations with a more realistic scenario.
Do you know what is the most popular CPU sold in the market? It is Intel X5600. That has typically 6 cores. The fact you used an AMD 12 core CPU is artificially describing a corner case: most customers (80%+) have 6 cores per socket. Try out your calculation with these knind of CPU and 128GB RAM and you will be have a better picture of reality.

How do the new vRAM entitlements affect you?

Please take 2 minutes of your time to fill out this vSphere 5 migration survey:
[url]http://wuffers.net/2011/07/18/vsphere-5-migration-survey[/url]

We need more data! Results will be posted in the main vSphere 5 licensing thread over at VMTN:
[url]http://communities.vmware.com/thread/320877[/url]

First round of results here: [url]http://communities.vmware.com/message/1795012#1795012[/url]

Hi Rynardt,

The discription is excellent and can be easily understood by any one. But I would like to add a note to your description that the licensing is still based on Per Processor.

You should always meet the minimum licensing per processor along with the cap on the vRAM.

I set up virtual environments for my customers for a living. I see a lot of my customers using 3 dual socket servers (Xeon 5600-series, so 6 cores per socket at most) with 48 to 96GB RAM and buying vSphere Essentials Plus (was ~US$3500 if memory serves (VMware took their vSphere4 pricing down)).

It would be nice to see a pricing comparison for that type of setup.