It wasn’t exactly a secret that VMware was planning to change to a licensing model based on a count of virtual machines rather than CPUs for the next version of vSphere, the virtualization and cloud suite that serves as the core of all its cloud and virtualization product strategies.
VMware executives described the new approach, with few details, in early June, as they fanned out to talk about vFabric 5, the Java-based application development platform from SpringSource, which VMware acquired in 2009.
At the time VMware executives said the company would be shifting to a licensing model based on the number of virtual machines on a host server, rather than the number of CPUs.
When VMware formally announced July 12 that the cost of vSphere 5 licenses would be based on the number of virtual machines and amount of memory devoted to them — a complete reversal of its previous CPU-centric pricing model — there was outrage from customers. The new model left many feeling as if VMware was punishing them for having built whole virtual infrastructures using servers with the smallest possible number of CPU cores and highest possible memory. Some were shocked at the reversal; others said the company was taking advantage of its dominant market position in virtualization to squeeze more money out of them.
In vSphere 4 and vSphere 4.1, users paid according to total number of cores in the CPUs in the host server. A server running vSphere 4.1 Standard on servers with 1 CPU with 8 cores and 24 GB of RAM would cost $795, compared to $995 with vSphere 5. A vSphere 5 Standard Edition server includes 24 GB of RAM and costs $995 for the basic license. (Click here for a PDF of the full vSphere v5 pricing model.)
vSphere 5 licenses are based on “vRAM:” the total amount of memory dedicated to each virtual machine, not the limits of the physical server on which they run, according to Tim Stephan, senior director of product marketing at VMware.
That cuts the financial connection between physical server and virtual, and encourages customers to pack as many VMs as possible into each server something most are trying to do anyway to make their virtual server infrastructures more efficient, according to James Staten, principal analyst of Forrester’s Infrastructure and Operations practice.
It’s a big switch from license costs based on CPU cores, but makes it easier to manage performance based on the resources in a whole data center, because it allows customers to treat memory in many servers as one large resource pool that can be distributed to VMs according to need, Staten says.
One angry customer modified screenshots of the vSphere license agreement by expanding the options “I agree” and “I do not agree” to include “I accept the terms in the license agreement but strenuously object to them.”
Another wrote “The Five Stages of VMware Licensing Grief” to describe his disappointment in the change and the amount of work it would take to avoid what many on Twitter referred to as the “VMtax.”
VMware spokespeople said the anger stemmed simply from surprise, not real concern over either new prices or the new emphasis on RAM rather than CPU cores.
Customers talked more about costs than novelty, however, building and posting utilities to help one another calculate the real budget impact. PowerCLI specialist Luc Dekens wrote a Powershell script to help other customers identify whether their VMware infrastructures are covered by the vRAM model.
VMware posted a questionnaire-based planner to help customers map out the most effective set of requirements for them under the new rules along with a script to validate existing licenses and the vRAM capacity to which theyd be entitled under the new model with no additional cost.
Forum guru Rynardt Spies posted a cost analysis and comparison of a single cluster of ESXi hosts using all of the various editions of vSphere. In all of them vSphere 5 comes out as significantly more expensive, but not enough to cause mass defections.
Spies and Tom Howarth from PlanetVM, who helped with the analysis, decided “[while] we are not entirely in favor of the new licensing model, it does seem to, in certain conditions, force you to stay with recommended best practice.”
Reducing the VMTax
Customers can take steps to minimize their costs under the new licensing model, even if those changes may not be easy to make, Staten says.
1. Stop Over-Committing Memory
The most obvious is to stop packing servers with as much memory as they can hold. Similarly, dont keep assigning more memory to VMs running certain applications just to boost performance a bit or make the application owners feel better about how much RAM they have available.
The new model creates a huge incentive for customers to analyze their actual resource use and distribute it efficiently, rather than throw memory at performance problems until they go away, Staten says.
2. Make Your Vision More Cloudy
The model for vSphere 5 is much more in line with cloud computing than with traditional physical servers or first- or second-phase P-to-V migrations, Stephan says.
That’s one reason VMware decided to treat all the vRAM in a virtualized infrastructure as one pool of resources, rather than separate puddles attached to specific physical servers.
Even vSphere licenses that are paid for but not deployed carry the right to use additional memory, so it makes more sense to calculate your available resources by counting licenses first and multiplying the amount of permissible vRAM than to count either VMs deployed or physical servers. Total cost is based on the amount of vRAM assigned and used by specific VMs, no matter where the physical memory chips live within the infrastructure.
Pricing VMware according to the resources accorded each VM makes more sense as customers move toward more cloud-based virtual infrastructures, which allow applications to use memory, storage and other resources from different parts of the data center, or to different data centers altogether, Stephan says.
3. Ramp Up Your Asset Management
vRAM entitlements (the amount allowed to each virtual machine and the total available to all VMs) are isolated according to the edition of vSphere for which they were purchased. A vRAM pool can include all the memory available to all the VMs running on all the hosts supporting vSphere Standard edition, for example.
Adding in a branch office that runs vSphere’s Essentials Plus edition doesn’t increase the total. vRAM assigned to Essentials Plus workloads has to stay with them, just as Enterprise Plus vRAM has to stay with VMs running that edition.
If all the servers in a cluster run the same edition of vSphere, however, the amount of vRAM that can be pooled is unrestricted. It is limited only by the amount of RAM actually installed.
Customers can add more vRAM to an infrastructure whose pool is maxed out at any time by buying additional licenses, according to VMware’s white paper guide to the new pricing model.
Employing VM lifecycle management tools becomes far more important under vShphere 5 licensing. Such tools limit virtual machine sprawl and the waste of computing resources by killing off unused VMs and limiting the lifespan of those that are rarely used. Since vSphere 5 costs are based on the amount of vRAM consumed, using less vRAM at any one time reduces the total cost.
Only vRAM assigned to virtual machines that are currently running counts against your total. Shutting down VMs that aren’t used at night frees up the vRAM that was assigned to them, potentially reducing the total cost of licensing. In that respect the pricing scheme is similar to concurrent licensing customers pay for the number of products they are running, not the number of machines or end users who have the potential to run them.
5. Analyze, Analyze, Analyze
vSphere 5 licensing costs are not retroactive to vSphere 4.1, even in environments that mix the two. It therefore pays to analyze where it makes sense to deploy vSphere 5 for the additional functions it provides.
Those functions include greater potential scalability of individual VMs, a new virtual machine format that handles 3D graphics more efficiently and the ability to run the VMware vCenter management application as a Linux-based appliance rather than installing it on one of the servers it is supposed to monitor.
Customers may actually be able to cut costs if their analysis of VM performance requirements are sharp enough and cost/benefit criteria are defined clearly enough, Spiers wrote.
Upgrading existing servers from v4 to v5 applies the new cost model; there is no grandfathering. Applications or servers that really need fat piles of memory may be best left with the older edition, or even shifted to physical servers rather than virtual ones running vSphere 5, Spies wrote.
“People can forget the hype about a VM with 1 TB allocated memory, as it would be much more cost effective to run that server as a physical host,” he wrote. “Allocating 1 TB vRAM to a VM will require you to have 22 Enterprise Plus licenses, and at a cost of almost $78,000.”
IT enterprise architect and vExpert Mark Vaughn compared a range of server scenarios to analyze costs in specific situations to give real-world examples of how the costs change according to specific configuration requirements.
His conclusion: Analyze how much memory each VM requires to run optimally, and inventory the amount of RAM available within the entire IT infrastructure. Mix and match to get the best performance, then buy more licenses if you really need to use all that memory.
The key is to see all the RAM an organization owns as a single, malleable resource, which Staten says has the potential to make IT’s use of its own resources far more efficient than it has traditionally been.