When VMware announced its new vRAM-based pricing, it was soon apparent that this model would discourage the use of hyper-dense memory servers like Cisco’s Unified Computing System and, more significantly, would discourage the very process VMware is trying to encourage--the virtualization of larger, mission critical servers. If virtualizing a 128-Gbyte Oracle database server meant buying two more licenses for vSphere Enterprise Plus, the savings we were expecting from virtualizing that server go straight into VMware’s pocket.
The math got extraordinarily out of whack when considered along with the boost in the maximum memory a virtual server can have to 1 Tbyte. That proverbial 1-Tbyte VM would have required $30,000 in additional vSphere licenses under the initial scheme but, under the new plan, a single VM will only hit the vRAM pool for 96 Gbytes or exactly one Enterprise Plus license.
Many bloggers, especially those with basement labs, also complained about the meager 8-Gbyte vRAM allotment for the free version of ESXi. Given that I had to upgrade my desktop to 8 Gbytes to handle the 347 tabs I sometimes have open in Google Chrome, 8 Gbytes was a bit too little. VMware has upped the allotment to 32 Gbytes, which is in line with the capacity of single-processor servers like HP ML110s or Dell R210s that populate most basement labs. Unlike the more commercial versions, this limit is on physical, not virtual, memory and is an enforced limit.
The last change, and one that could end up having a big impact, is that VMware will no longer use the peak vRAM utilization but the one-year average for licensing. In the old system, users running parallel systems for a few weeks or large test and development images for a few days may have found that they exceeded their limits and would have had to buy more licenses. With averaging, those situations won’t cost any more. Since many organizations run servers that are only used by office folks that are in 9 to 5 and load-balanced web or application servers, I wouldn’t be surprised to see Veeam or VKernel come up with utilities that power servers up and down on a schedule. After all, just shutting down the MRO server when the plant is closed could cut its vRAM usage by 50%.
Given the vocal reactions when other vendors have shifted to more usage-based pricing, I can’t help but think that, while the community of VMware bloggers and tweeters are happy that VMware listened and increased the allocations, I’m pretty sure those self same folks would have been writing about how awful it was that VMware was shifting to charge for vRAM if VMware had announced the new pricing model with the higher allocations. I don’t think anyone at VMware was wily enough to announce the lower allocations as a strawman, knowing people would be so pleased when the limits were lifted that they’d forget they were upset there were limits in the first place. Of course, if I’m wrong, that guy deserves a raise.