By some of today’s standards, we’ve got a modest sized virtual infrastructure. Seven non-production ESX hosts and four production ESX hosts, plus various isolated lab and small remote office deployments. We purchased the infrastructure a few years ago when we stood up our first ESX environment.
From day one, our infrastructure has been plenty over capacity because part of our initial deployment didn’t involve a massive P2V exercise to immediately fill up the infrastructure. The goal was to gradually migrate new and existing virtualization candidates to the virtual environment. With around 90 VMs today, we don’t quite have the consolidation ratio that I’d like to be seeing, but I understand it’s a gradual ramp up and for the most part I’ve been patient.
Like others, we began experimenting with VDI (virtual desktop infrastructure) on VMware ESX. A VDI image was developed and deployed to 12 pilot users. For the past year, the pilot testing has been largely successful. In fact, the tales of success in the hallways produced a few additional requests from developers and power users for VDIs. Today we’ve got 20 VDIs with a batch request in the queue for 90 more.
I was excited when I heard 90 VDIs had been requested. In a single transaction I could double my consolidation ratio from 8:1 (virtual machines:physical machines) to 16:1. VM to socket ratio would be 4:1. VM to core ratio would be 2:1.
Disappointment soon set in. Management has put an indefinite hold on VDI deployments because we don’t have a chargeback model that can be applied effectively or fairly to VDIs. The primitive chargeback model we’re required to abide by comes from the Finance department. It states that in order for us to charge a business line, we need a vendor invoice. Basically, we can’t charge entities for use of existing infrastructure. We would need an invoice for infrastructure expansion such as additional processors, disk shelves, memory, fiber switches, etc. What it boils down to is we’d end up charging a business line thousands of dollars, sometimes even $20,000 or more, for a VM or a few VMs. The business line immediately walks away from the table and decides to purchase a few comodity PCs for $500 or less each. In addition, the business line learns that all of this virtualization money saving that they have been hearing about is totally false. They get a bad taste in their mouths about virtualization and spread their experience to other co-workers and departments.
The technology department is not a profit generating department. We’re completely expense. We don’t have the money in our own budgets to fund additional virtual infrastructure without a project. Budget requests have been submitted to fund VDI infrastructure, however, the red pens see that as way too much expense when commodity PCs can be purchased for less. Particularly in today’s business climate (and we’re a bank). A lot of the tangible benefits that virtualization brings to the datacenter (cooling, energy, real estate, etc.) aren’t directly realized out on the floor where the desktops are. A manager purchasing five PCs for his/her department doesn’t have cooling, electrical, or space issues so a $1,000+ per VDI cost makes no economical sense to the departmental budget they manage, even though it could be proven beneficial at a higher level for the company.
Organizations must upgrade their chargeback models and cut out the politics to allow virtualization to grow. I think it’s safe to say that virtualization isn’t going away anytime soon. The trend is here to stay. Adapt now or adapt later. The choice on how soon we’d like to save money and the environment is a decision that all must be in agreement with to move forward.