The calm after the storm – My take on vRAM

** I have modified this post to include the updated licensing changes announced 8-3-2011, as well as the link to Alan’s updated PowerShell script.  The video clip has not been updated, the concept is the same only the number have changed.  Enjoy!


***Disclaimer – I am a VMware employee and receive paychecks that have the word VMware stamped on them.  My thoughts are my own but if you are afraid of an employee’s opinion, run away now!



A lot of big announcements have taken place over the past week from VMware.  We had our largest launch event in history and announced to the world that our flagship product “vSphere 5“ is being released.  We also announced many other product updates and releases like SRM 5, vShield 5,  vCD 1.5 and an updated release of heartbeat for Virtual Center.  Loads and loads of exciting new features and functionality to allow you to run your mission critical workloads on the worlds #1 hypervisor.

I wanted to do a factual write-up on what has changed, hopefully explain it better since there seems to be a lot of misconceptions, and also give you my take on what I am seeing from my customer base.  For those of you that don’t know my background, I came from a customer environment where I designed and implemented VMware in a large scale deployment.  I then decided to convert to the dark side and go work for a vendor (VMware) so I actually have some perspective to offer from both sides of the fence.  Now that we are one week out from the new licensing change, I thought I would share some thoughts.


What is vRAM?

One of the changes that came along with a lot of exciting new features was the new vRAM licensing model.  VMware has decided to move away form the core based model to a “consumed virtual ram” model across the entire environment.  For those of you that are unsure of how the current vSphere 4.x licensing model works today, I have embedded a chart here to help you understand the core limitations as well as the features and functionality between the versions.




Below is the new vSphere 5 pricing comparison which also includes a feature/functionality breakout.  As you can see it lists the new vRAM entitlements and how much pooled vRAM you are allotted  per socket.  Notice we have removed the core limitations around the physical processors, and lifted the memory limitation from a physical perspective.



Your current licenses will be converted to the correct vRAM allocation model depending on the version of vSphere that you are paying maintenance on.   VMware has provided a power shell script to assist you with determining the amount of vRAM you have available to you when you decide to upgrade your environment.  My suggestion is to download the power shell script from VMware’s Alan Renouf that will automatically calculate this information for you.

I decided to put together a quick video link that discusses vRAM in a little more detail, and gives you an example of what this pooled model looks like across a 3 node cluster.  Watch the video below:



My thoughts

There has been a lot of emotional responses around this topic over the past week, which is understandable.  VMware has the best user community of any software company I have ever seen.  There are several reasons for this in my opinion.  VMware has made great impacts on our customer environments from a datacenter consolidation perspective.  We have allowed our customers to run more efficiently, do more with less, and enabled them to become hero’s for their internal customers by offering them agility.  We have also given them a portion of their lives back from an administrative perspective. 

I was one of these end users that came in to fix systems at all hours of the night when the hardware went south.  I was responsible for deploying hundreds and hundreds of physical servers that eventually consumed all of my time from a maintenance perspective.  VMware technology gave me my personal life back, and allowed me to start doing mundane maintenance tasks during the day!  Storage vMotion allowed me to retire 3 older storage array’s (30+ TB) that were at the end of life, and move them to newer technology with no downtime,  during the middle of the day!  I believe there is a personal component to all of this, and part of this is what makes the community such a strong force.

The goal of this article is to educate people, and give you my perspective.  It is not intended to defend VMware’s position or try and convince you why vRAM should make you sleep easier at night.  VMware has spent the past two years working internally and externally with customers to try and determine a fair licensing model that works for everyone.  The current model will not scale with the quickly changing hardware landscape driving core counts exponentially.  A handful of my customers are already having to double up on licenses as they are at vSphere Enterprise and are being restricted on the number of core’s.  I think the model is a fair one and as you walk through it hopefully you will see the logic behind why VMware had to make this change, to better support our users and the community.

  1. The model idea is almost fair, but not the price! VMware is playing dangerous game trying to convince customers that vRAM licensing is good, but what they fail to mention every time I read is that you have to license new server with physical CPUs no matter what, even if you do not even need more vRAM into the pool. With new licensing price it is no brainier for CFO to push for evaluation of other hypervizor. % of of customers will find other hypervizor good enough for portion of thier non critical workloads. VMware has to reconsider the price or how everyone is calling it vRAM tax, not vRAM model itself. 

    Why would you confuse a customer by mandating to purchase 2 licenses for 2 socket server? What if that customer only plan to run one 4G VM for start? Is this a fair licensing? What do you think?  

  2. Thank you for the video. However, I think your assumptions of 64GB per host is VERY light.  Our 2 year old cluster (5 hosts) had 96GB RAM across 2 sockets (4 cores/socket).  This cluster started out with 48GB RAM.  Our new cluster we just purchased is 256GB per host.  Our current cluster will move to our DR site.  We aren’t a huge shop by VMware standards.  We purchased the Enterprise+ license when it was released for the features.  I don’t know of any shops of our size that are this light on RAM.  

    While we are still under our vRAM pool limit, we are short licenses as we approach the max configuration of our cluster (80% X N) using an N+1 cluster.  

    The vRAM tax is real. It’s not a matter of IF we will get hit, it’s a matter of WHEN.  Even VMware’s own numbers predict as high as 20% of customers will be affected.

  3. The survey is okay, but it is not real data is it… it is the perception of someone on his environment not the actual facts. Reality is that we all think we have a much higher consolidation ratio than we actually have because of things like HA N+2 etc.

  4. Agreed Duncan, I have been getting a lot of Alan’s surveys back and the true data is in the results. I am going to try and tally up all the data and post the results across the board. This might take me some time to do though…

    Thanks for your feedback. Your getting great consolidation ratios, I think if you run the ROI model even under the new vRAM model, it will show you the value of VMware. We offer a superior product and help you transform your datacenter like no other company. Thanks for your business.

    I will disagree on the cost and vTax argument along the same lines. VMware had to change the model for the long haul to something that is not based on physical limitations. Change is never easy when software companies have to impose new concepts like this, but honestly we have been working on this model for over two years and are trying to do what’s right for both you the customer and VMware. Along the lines of other evaluating other Hypervisors, we certainly hope you see the value of the technology you have been using for years, and hope to retain you as a loyal customer. I understand if you are getting pressured to look at other technologies for dev/test systems. Hopefully you can highlight all the great features we are delivering back to the business and show the ROI model to your CFO. Thanks for your business.


  5. my thought is simple. VMware controls the hypervisor. instead of billing based on how much memory is PROVISIONED, bill based on how much is USED. This data should be easy to get at. And it should be real data, e.g. if my linux guest OS is using 16GB of memory for buffers that should not count because that memory is not used by applications(and I have no way to control how much memory the OS uses for buffers/cache).

    there should also be a separate license for vRAM, at a significant discount to the CPU+vRAM license. Installing 4-8x the number of CPU licenses required just to cover the amount of memory installed is stupid. It encourages the use of dual socket systems with 48GB of ram/socket. Bring back 4GB dimms and 1GbE..

Comments are closed.