Anyone that knows me or has worked with me will know I like to stay at the technology end of the business and will happily leave the others to worry about the licensing and cost. With VMware’s latest release they are making this somewhat more difficult for me as we are now licensed on not only physical CPU’s but also allocated memory. I thought I would put together a blog post to explain the new licensing model in some detail and what this will mean generally for the customers I work with.
Please note a good place to start when trying to understand the new licensing models yourself is the VMware vSphere 5.0 Licensing, Pricing and Packaging white paper
vSphere 4 Licensing
Currently in vSphere 4 we have 6 different levels of licensing, these are as follows
What this has meant in recent years is that the amount of RAM allocated to ESX(i) hosts has been increasing alongside consolidation ratios without the customer needing to purchase any additional licenses from VMware, so increasingly with the higher core counts available today we have seen in some cases that actual amount of physical hosts going down along with the amount of VMware licensing required to run the same amount of virtual machines. For this reason it was to be expected that VMware would be making some form of change to the licensing model.
vSphere 5 Licensing
So moving on to yesterdays announcements and the change in licensing for vSphere 5, VMware are moving to a model where you are no longer restricted to the amount of cores each version supports but to a model where it is still licensed on a per physical CPU basis, but are also licensed on the amount of allocated vRAM across your pooled resource.
The first part to note is that the Advanced licensing is now end of life and actually managed to out live Enterprise that was supposed to be retiring, anyone with an existing advanced license with support and subscriptions will be upgraded to Enterprise. An important part to understand for the new licensing model is the amount of vRAM allocated is calculated across the pooled resource, this means over your vCenter and more importantly can also mean vCenter’s linked with Linked Mode. This means you are only having to license what is actually being allocated in total to your virtual machines and not what amount of memory the physical host contain and equally and more importantly if you have a DR environment that is largely sat there dormant the vRAM entitlement that you get with these can actually be used by the hosts in your production clusters.
Let’s look into a few scenarios and see how this end’s up working based on some end users.
We will start with Essential Plus as my guess this will be the most popular of the essentials bundles.
With your average two way ESXi Host (HP DL360 / 380, Dell R610 / 710 etc)
With this licensing level and the 3 host maximum you would be entitled to 144GB (24GB x 6) of allocated RAM for all our virtual machines in your clusters. In a N+1 configuration and taking a maximum figure of 4GB’s allocated for each of your virtual machines this would allow you to run 36 servers in this configuration, in a real life scenario this would probably be more between 36 and 72 right sized VM’s.
From a point of view of all of my customers with Essentials Plus I don’t think any of them have an environment today that would not fit within this new licensing model.
The majority of my customer in the next tier of licensing generally have Advanced or Enterprise licensing so I will concentrate on looking at the Enterprise licensing next. Again I will concentrate on your average two way ESXi Host (HP DL360 / 380, Dell R610 / 710 etc)
With this licensing level and a cluster of 7 hosts you would be entitled as standard to 448GB (32GB x 14) of allocated RAM. In a N+1 configuration and taking a maximum figure of 4GB’s allocated for each of your virtual machines this would allow you to run 112 servers in this configuration, but again in a real life scenario this is more likely to be somewhere between, 112 and 224 right sized VM’s.
For the majority of my customer in this space this would also more than meet their existing requirements, the few customers that I do have that this doesn’t potentially meet their requirements for also have another aspect to look into. These larger customers tend to have a DR environment that is largely (Not completely) sat dormant and connected to the production vCenter using linked mode. This means that potentially they have another 7 hosts licensed and sat in a DR datacenter not generally being pushed during the working day. For these calculations I will assume that 1/2 of the physically available resource is allocated to test and dev outside of a DR event. So based on that these customers have available to them a further 224GB of vRAM that will be in the pool available for the production servers to utilise. I now believe that all of my customers requirements are met with standard allocations of vRAM.
Time will tell as I start discussing the new licensing model with my customers if these calculations actually do make sense for them and I imagine a few may still end up having to increase their licensing from Enterprise to Enterprise + to get the further vRAM allowance or purchasing additional Enterprise licenses to allow them to allocate vRAM,
Points to remember to doing the math
* vRAM does not equal pRAM (physical RAM)
* Consider your DR environments licensing in your pooled resource
* This is probably the right time to ensure that your VM’s are right sized (Does that VM really need 4, 8, 16,24 or more GB of RAM!?)
* This is a good time to review chargeback and ensure the people requesting more RAM in their server understand the cost.
That is my considered thoughts on the new form of licensing, it is never going to be popular VMware restricting the licensing in anyway but I think we can all see why it had to be done, I am sure there are going to be many more Enterprise customers effected with their larger environments but from a quick look at the SMB / SME space things don’t look too bad from here. I would be pleased to hear your comments and will update the post with any new findings etc.
Something that has been bought to my attention is what happens when you need to do a DR test, will this mean you go over your licensed limit. I will see if I can get some clarification on this from someone at VMware.
To confirm the limits are soft limits, so theoretically you could still complete your DR failover test I need to confirm with VMware if this is acceptable in terms of the license agreement though.
Barry, thank you for being rational about this and doing the actual calculations. Far too much uneducated ranting on Twitter at the moment – obscures the situation.
There might be an exception for this in the licensing docs that I’ve missed, but you say you’d consider DR site vRAM as part of the licence pool, which is fair enough.
However, a huge benefit of SRM is the fact that you can run a DR test any time you want without shutting down live systems – when you implement the SRM test, you’re going to fire up a bunch of VMs and breach your vRAM limit aren’t you?
Here’s a script to help you calculate license requirements for vSphere 5:
Fantastic many thanks Hugo
Hi Barry – Our VMWare Account Rep told us today the vRAM limit is a soft limit. Basically it’ll get reported in a log file that VMWare can access if they ever need/ want to.
Realistically, he said if you go over once a month, it shouldn’t be an issue. If you are constantly over, then it’ll mean buying more licences. But of course that’ll all depend on your customer relationship as well.
I know I am downloding XEN and Hyper-V to have a play with tonight. In our particular environment, the cost is going to be prohibitive come upgrade time…
This new model is quite “hard to sell” to customers. Wouldnt it be less greedy and more rational to licence per “real physical RAM consumed” ?
Imagine your car dealer tell you – you can drive with 3 gallons gas per 100 mile with this car – but when you buy it – we have a deal with the gas stations that you have to pay the double price for gas.
In a model where i rent the license in a flexible way i would fully support a “per assigned resource” pricing system maybe.
You (at least me) tell the customer who came up with comparing benchmarks “How many RAM and CPU did they overbook ? VMWare makes more from less.” and now you have to tell them “Off course you can rely on transparent page sharing, great CPU scheduler, etc .. but you have to buy then licenses for resources you don’t own in HW.”
The “twice your RAM in a DR environment” statement is stupid – who wouldn’t fill up both environments with less critical systems and would shut them down in case of a DR event – this is promoted practice by SRM.
Generally I don’t see the dr enviroment “filled up” with less important machines hence the reason I took te figure of 50% into consideration. I generally do see the DR enviroment used for test and dev or lower importance machines but generally not maxed out. Many thanks for your opinions though.
I move to Australia, look into the vRAM changes and your page pops up 🙂 genius, cheers fella
Glad to hear you finally got out there! How’s it all going?