vSphere 5 Licensing Model – Death to Scaling Up?
Update 08/03/2011:
With the changes to vRAM (essentially doubling the amount of vRAM entitled per license) the post below no longer has applicable math – my entire environment is now fully covered by vRAM licensing. However, I am going to leave the post intact for historical purposes so later on we can all look back at how fun this has been. Now that I know I can afford to upgrade, I can all finally give 100% of my attention to the cool features in vSphere 5!
Now, back to the original post:
Amidst all the fun and geeky excitement of the July 12th announcement of vSphere 5, there was also the announcement of a new pricing model for vSphere host licensing. I’ll admit that this was not one of the things I was looking for this morning; I was busy digging into Frank and Duncan’s new vSphere 5.0 Deepdive book, absorbing news on the new enhancements, and getting my “tweet on” with fellow VMware buds. Now that the dust has settled and I’ve had some time to absorb the pricing change, I thought it would be good to have a look at my environments and how they would be impacted.
This isn’t a gripe session or meant to spit on anyone, but simply a look at one customer’s environment (being myself) and how this change will affect things with the information that is currently available. I’m not looking to start a flame war here.
I run a modest pair of datacenters, each with it’s own cluster.
Memory disclaimer: The environments that I oversee will (and in some cases do) consume the majority of the memory presented. I understand that the new vRAM licensing is a consumption based model, and that in the case where memory is not being utilized by VMs on one host it can be distributed to VMs on other hosts. However, I’m in the position where any idle capacity is snatched up readily. If you have idle memory sitting around in your environment, I’m jealous!
Datacenter A:
Current Setup:
- 4 Hosts (HA Cluster)
- 4 CPU sockets each = 16 CPUs
- 512 GB of memory each = roughly 1536 GB of memory available for VMs (N – 1 for HA)
- Using 16 sockets of vSphere 4 Enterprise Plus licensing
Upgrading to vSphere 5 Enterprise Plus:
- Would entitle me to use 48GB of memory per license on VMs.
- 48 x 16 = 768 GB of vRAM licensed for VMs.
- To achieve 1536 GB for VMs = 32 vSphere 5 Enterprise Plus licenses required in an HA config (16 additional)
Datacenter B:
Current Setup:
- 7 Hosts (HA Cluster)
- 2 CPU sockets each = 14 CPUs
- 192 GB of memory each = roughly 1152 GB of memory available for VMs (N – 1 for HA)
- Using 14 sockets of vSphere 4 Enterprise Plus licensing
Upgrading to vSphere 5 Enterprise Plus:
- Would entitle me to use 48GB of memory per license on VMs.
- 48 x 14 = 672 GB of vRAM licensed for VMs.
- To achieve 1152 GB for VMs = 24 vSphere 5 Enterprise Plus licenses required in an HA config (10 additional)
Total Datacenter Licensing Requirements for vSphere 5
Licenses Required to utilize maximum memory for VMs (in an HA config): 56
Licenses purchased with vSphere 4: 30
Net difference: 26 additional vSphere 5 Enterprise Plus licenses required (86.7% increase)
Update: Real World Numbers
Using the vRAM entitlement PowerCLI script, my actual licensing requirement at this time is as follows:
It’s too early to tell exactly how I feel about this licensing change. I’m happy that SnS will cover the 30 current licenses of vSphere 5, but justifying 26 more licenses is a bit tough to swallow. This is a steep additional licensing cost to enter the vSphere 5 era. Is it worth it, for all those new features (Storage DRS, the new HA, Auto Deploy, etc. etc.?). Jury’s out for now. I will admit that I want to use vSphere 5!
I do agree with a friend of mine who stated that “those that scaled up heavily” are up a creek. Even in the vSphere Design book, the question of scale up vs scale out is addressed as more of a technical and risk challenge. Is it now more of a licensing challenge?
Design decisions can fan in from a multitude of sources. Licensing is no different conceptually.
I look at this as just another design decision where justification should be made and impacts identified (whether positive or negative). Initial indicators are that this new licensing model has an impact of higher cost for scale up environments. Conversely, know the impacts in the design decision of scaling out. That’s where we all just came from in the physical world. Datacenter buildouts. Real estate costs. Rack space. Heat. Cooling. Cabling. Depreciation. etc.
Thanks Jason, very honored to have a comment from you on my blog.
I fully agree that licensing is a design consideration – it has worked in my favor by scaling up in the past (cheaper to stick in more RAM in a rack server than buy another host with more vSphere 4 licenses, although risk is higher with larger consolidation ratios).
I think right now it’s a bit of sticker shock for me to see how many additional licenses I’ll need to cover my VMs, and I’m worried about how I’ll get justification from the business for the required funds to procure the licenses.
Scaling out incurs a different licensing tax – namely Microsoft Windows Server Datacenter or RHEL per-socket licenses, as well as any 3rd party tools you may use (Quest, Veeam, vKernel) all of which are usually licensed per-socket (this may change now).
Due to the changes with vSphere 5, we’re now forced to be extra skimpy with initial allocations of memory (rather than throwing physical RAM at the hosts and not fighting that political battle). At least with the latest O/S’s (Win2k8, etc) we can dynamically hot-add memory without rebooting (whether the app detects the change is another story).
The VMs that just became very expensive are those that had large amounts of cache (databases, web application servers, messaging servers).
My 2¢
Those are some good considerations to keep in mind for scaling out vs up. I’ve typically shot for around an 8:1 Memory to CPU logical core ratio (hexcores with hyperthread), as it works very well in my environment’s VM load.
My math also doesn’t assume anything like overhead (both Host and VM) and is just a raw number, but in any event, I’m going to need more licensing.
Here’s a script to help you calculate license requirements for vSphere 5:
http://www.peetersonline.nl/index.php/vmware/calculate-vsphere-5-licenses-with-powershell/
Hugo
Chris, you’ve got a good discussion going here, so I don’t want to nitpick, but since it does come down to the numbers — for this theoretical cluster, are you running all 6 machines at 100% CPU utilization? Remember vRAM != pRAM. Also, in your environment, do you have other clusters or machines that are not as heavily utilized that could add to the vRAM pool?
This new licensing *is* intimately connected with your *total* real world design. Calculations on isolated physical components may not proportionately reflect the total cost. Already your cost with 6+1 cluster design looks different from looking at a single server.
Don’t worry John I don’t think your nitpicking. Part of making this post was to get feedback from others on my situation so I can sort this all out.
Regarding 100% CPU utilization, none of my hosts get close to that number. I commonly see it average in the 30-50% CPU utilization range.
In case you were talking about memory, I tend to keep each host at or below 85% memory utilization, as that is my ideal “HA reserve” value (1/7 = 14% of a cluster). I do have some unused memory in my 4 node cluster that I could borrow from, but that will be consumed as P2V projects are completed. Most of the memory is vouched for; net new growth results in me purchasing additional servers (it doesn’t typically sit idle).
I don’t think it’s typical for a design to have significant amounts of unused memory. I’ve always considered memory to be the quickest consumed resource, and the one that most designs focus on first (not that the other resources aren’t important). Rarely do I see CPU consumption at a value worth concern (CPU ready is another beast all together).
I will say that at the end of the day, do I absolutely need 26 more licenses? No. There’s things like overhead that will eat into my available number. Will I need more licenses to cover the VMs I have –right now– ? Yes, definitely. I’m not comfortable posting how many VMs I run and how much memory they consume (I feel this is corporate confidential information), but I can say that just based on raw consumed VM guest memory (where a VM given 4GB of memory = 4GB of vRAM needed) I will need many more than 30 licenses of vSphere 5 Ent Plus.
Woops, I meant pRAM utilization. Looking at your results over on the thread at communities and *currently configured* vRAM (not theoretical maximums), it does look like you need an additional 16 licenses. (Am I reading that right? I don’t think the 26 license result above is that meaningful, since you aren’t going to run at 100% pRAM).
Any thoughts on how your cluster or VM designs could change (now or in future), or (and at this point I don’t want to be insulting
any spare DR or low-utilization clusters you could put into your pool?
I figured, so I tried to answer both versions of the question.
Yes, I am in current need of 16 more licenses based on the community thread post and using the PowerCLI script to properly license my VMs (I’ll update this post with the results). The reason for this is that my 4 node cluster still has capacity remaining; as planned growth occurs I will consume more vRAM on that cluster. Probably not the full 26 more, but I will point out that my math above does assume N+1 (HA) configuration, and I removed an entire host’s worth of memory from each cluster’s usable vRAM. For the 4 node cluster, this assumes each host will consume 75% of it’s phsyical RAM, not 100%. For the 7 node cluster, I’m assuming 85% pRAM used per host. I don’t think 75% or 85% of consumed pRAM is an unrealistic number on a VMware host – after all, they are great at utilizing pRAM.
No insults taken: I don’t have any other clusters that would help the math, such as a DR site or low-utilization clusters, to put into the pool. I purposely left off my VDI environment (XenDesktop on vSphere 4) as it would significantly drive up the amount of licenses I will need (these hosts are 4 pCPUs with 768GB of pRAM). I’m strictly keeping the numbers to my vSphere “Server” clusters.
I’m not entirely sure how this will affect future design, I will definitely think harder on the procurement of 2 socket 192 GB hosts. Scaling out comes with many costs prior to vSphere 5: soft costs like Microsoft licensing, monitoring socket licenses, configuration; and hard costs like physical space, wiring, racking, space on the (expensive) switches, etc. Scaling up was a good way to mitigate those costs.
How do the new vRAM entitlements affect you?
Please take 2 minutes of your time to fill out this vSphere 5 migration survey:
http://wuffers.net/2011/07/18/vsphere-5-migration-survey
We need more data! Results will be posted in the main vSphere 5 licensing thread over at VMTN:
http://communities.vmware.com/thread/320877
First round of results here: http://communities.vmware.com/message/1795012#1795012
Dear all… the information i have from VM site is that it supports 96 GB per CPU with Enterprise Plus…
http://www.vmware.com/files/pdf/vsphere_pricing.pdf
I am not sure where is 48 GB per CPU mentioned… can you please refer to any official link.
This article was written when the vRAM announcement was first released, and at that time it was 48 GB of vRAM for each Enterprise Plus license. It was later doubled to 96 GB.