For simplicity, we will use VDI to refer to on-premises deployments and DaaS to refer to all-cloud deployments.
On-premises VDI is highly over-provisioned. According to studies conducted by IBM, McKinsey, and the National Defense Research Council, a typical data center is at 12-18% utilization. This is not surprising because infrastructure is provisioned for peak usage, which may only happen for a few hours a day. And the infrastructure sits idle or significantly under-utilized the rest of the time. Furthermore, it is difficult to assign cost to usage. Your infrastructure might be used by different kinds of users — software developers, contractors, call center employees, and others. Their infrastructure requirements and usage patterns are widely different. Can you bill the business unit for the infrastructure their use case requires?
With DaaS you can radically optimize your costs in 3 ways:
- Just-in-time Provisioning: Why pay for capacity when you don’t need it? Provision the machines just when they are needed and shut them off when they are not.
- Just-enough Provisioning: Since provisioning desktops can take 1-5 minutes depending on the cloud and type of desktops required, you can reduce connection times for users by provisioning desktops in anticipation of users logging in, e.g., shift changes, etc. There is a danger of over-provisioning if desktops are provisioned too far in advance. With Machine Learning, you can predict user login times, and provision just enough to keep user connection times optimized.
- Fine-tune Cost to Usage per user: We see a lot of variation in number of hours of usage per user. The cloud allows you to match cost to usage on a per user basis. For example, a user may use a GPU workstation for 2 hours a day at $1/hr, a call center agent may use a CPU desktop for 8 hours a day at $0.20/hr, and a software developer may use a large CPU machine at a flat rate per month. Now you can bill the business unit for cost based on usage vs. some fraction of an opaque infrastructure budget.
These cost optimizations are only possible with DaaS and will allow you to align cost with usage on a per user basis. Now, compare this to an over-provisioned infrastructure model you have with VDI where it is difficult to align cost with value. See the significant difference?
