VMware backs down on vSphere 5 virtual memory tax

Big Blue monopoly lesson

VMware has adjusted the price of its upcoming vSphere 5 hypervisor, responding to criticism from customers.

The server virtualization juggernaut has learned a lesson about pricing that nearly killed IBM in the mainframe market in the 1980s and did kill Sun Microsystems in the 2000s: if you have a monopoly, you can't push the prices up too hard. Users won't just balk. They will take their business elsewhere.

In mid-July, VMware launched the ESXi 5 server virtualization hypervisor and related vSphere 5 tools for doing useful things with it such as VMotion live migration and distributed resource scheduling across a cluster, just to name two. ESXi 5 was boosted to run across machines with up to 2TB of physical memory (double the previous release), supported disks with more than 2TB of raw capacity, and could host as many as 512 VMs on a single physical server (up from 320 on the prior ESXi 4.X hypervisors).

The guest VMs on top of ESXi 5 were expanded so they could address as much as 1TB of virtual memory, virtual network bandwidth was pushed as high as 36Gb/sec, and virtual disk bandwidth was cranked up to 1 million I/O operations per second. This was all more or less expected.

What was not expected – and what VMware's users objected to mightily – was the company's decision to tie a license to an edition of the vSphere 5 stack to a specific virtual memory capacity, thereby forcing customers who have fat machines and fat VMs to potentially pay a lot more for vSphere 5. VMware took off the core caps it had on vSphere 4.X licenses, but put on memory caps, justifying this as a simpler pricing model and one justified by the functionality in the release and the use cases in the VMware customer base.

The reaction from the customer base – as seen in comments here at The Register and on other blogs – was swift and almost universally bad. Even allowing customers to pool virtual memory across hypervisors using the same vSphere 5 edition would mean, in some cases, that customers would pay a lot more for vSphere 5 than they did for vSphere 4.

To VMware's credit, unlike IBM in the late 1980s and Sun in the early 2000s, when their price lists for mainframes and Sparc servers got way ahead of their egos (which was saying something), VMware has gone back to the drawing board and made the virtual memory pricing more reasonable than it was on announcement day.

As explained in a blog post published Wednesday evening by Bogomil Balkansky, vice president of product marketing for virtualization and cloud platforms at VMware, the company tweaked its virtual memory pricing scheme a bit to cushion the vRAM pricing blow and to quiet the criticism ahead of vSphere 5 shipments sometime later in the third quarter. (The company did not make executives available for comment on the vRAM price change).

With the vRAM memory capping, each license for vSphere has a cap set at the edition level. If you exceed that vRAM allotment on the machine running the hypervisor, you have to buy a second, third, fourth, and so on license to vSphere to cover the virtual memory that the single hypervisor soles out to guest VMs. (It is important to realize you are not actually running multiple hypervisors on the machine, but just activating memory allocation capability).

The ESXi 5 VM monster is a little less scary

Under the tweaked pricing announced today by VMware, the entry Essentials, Essentials Plus, and Standard editions of vSphere 5 have their vRAM caps boosted from 24GB to 32GB. For companies running these editions on fat machines with gobs of virtual memory on the VMs, this amounts to a 33 per cent price cut.

The vSphere 5 Enterprise edition has a vRAM cap of 64GB and vSphere Enterprise Plus has a cap of 96GB, which is double the memory ceiling before and amounts to a price cut of 50 percent on heavily configured machines with fat VMs. And, perhaps most importantly, on the Enterprise Plus edition, 96GB is the absolute cap on the machine.

The net effect of this change is very significant for companies that want to deploy database, email, ERP, and other heavy workloads on top of the ESXi 5 hypervisor. As we did back when vSphere 5 was announced, let's take the extreme case: a Xeon E7 machine with 80 cores and 2TB of memory, and let's assume that it is running Enterprise Plus and all of the physical memory was allocated as virtual memory in a 1:1 ratio. (You can overcommit virtual memory and actually do more than that, but let's not go crazy with the complexity here).

In this case, vSphere 5 Enterprise Plus would have cost $150,285 – that's eight licenses to cover the eight sockets in the sever with 48GB each plus another 35 licenses to boost the memory to 2TB. After these changes, vSphere 5 just costs $27,960 - exactly what vSphere 4 cost with half as much physical and virtual memory supported.

Suddenly, this sounds like a deal. And obviously VMware has learned something important: Virtual infrastructure is on a Moore's Law pricing curve, too, and is not somehow magically exempt.

Another big change with vSphere pricing, and one that is hard to quantify without getting really precise in the examples, is that VMware will now measure vRAM usage on VMs on individual machines and in pools of machines on a 12-month average instead of a high water mark that only looks at spikes. The net effect is that you pay for licensed based on average vRAM usage over a long haul, not for peak capacity needs.

Finally, the freebie ESXi 5 hypervisor had its vRAM cap moved from 8GB up to 32GB, by the way, making it a lot more useful. ®