YAvLP – Yet Another vSphere Licensing Post

I was recently asked to write up my thoughts on the new vSphere5 licensing. In our current environment we have approximately 56 physical processors and 1536GB of memory spread across 18 servers. This infrastructure is supporting close to 500 virtual guests.

My view of the new licensing is the same as mainframes are licensed. You lease a large amount of equipment and when you reach your maximum, you call up your friendly neighborhood $vendor and ask for new license key that unlocks more of the hardware that is already on-site.

The only difference is that VMware stormed the market by removing the big hardware lease and using commodity server grade hardware. This caused all the server vendors to innovate and pack as much memory into a server chassis as possible resulting in fewer processor sockets and less revenue for VMware. Something had to be done.

Our Infrastructure

In doing my research, I ran one of the many available vSphere licensing estimation scripts. I choose the one written by Hugo Peeters since it did not require vSphere 4.1 (we are still on 4.0U1). The result showed that if we licensed each of our 56 physical processors, we would still have 330GB of available vRAM. The reason for this is because when I spec’d out our current hardware, Dell R905  with 128GB of RAM, I did not feel comfortable losing close to 150 virtual servers if a single host failed. We currently get 75 – 80 virtual guests per vSphere server.

The problem comes in when you look at our new flexpods; 6 blades with 2 physical processors and 192GB of RAM! We would need to buy 2 extra licenses for just one of these blades.


This licensing change is not a huge surprise, however I think VMware could have made some changes. While I do like that enterprise customers will not be hindered if they go over the vRAM allotment, VMware should drop the per physical processor licensing; Just sell blocks of 48GB vRAM and make a vCenter plugin to track and alert administrator to current total usage.

This entry was posted in Hypervisor. Bookmark the permalink.