On-Prem vs Cloud: Cost vs Uptime

Over the past two weeks, I’ve had several existing and potential customers talk to me about moving on-prem systems to the cloud, and pulling cloud systems back to on-prem. I want to talk a little bit about what the two major forces are that are pushing and pulling on these decisions.

On-Prem: A Lack of Redundancy

Although the details vary, all of our on-prem customers seem to have the same complaint: when something fails, it takes hours to get it fixed. Few customers buy more on-prem hardware than they need, and perhaps have one or two backup systems for critical data, such as accounting. Almost no heed is given to back services; having alternate file servers, web servers, or application servers when the primary encounters an issue that takes it out of commission. The disadvantage, of course, is that having a service backup is usually around twice the cost of having just the primary. And while economies of scale do apply, it’s a hard sell to justify a “just in case” server, disk array, or other network equipment. As a result, when there is a failure, usually something goes down for a least a little while, and a third of the time that will be during the working day.

Although Cloud systems are not immune to outages, recovery is usually faster. Many Cloud systems force users and administrators into a paradigm that is easier to recover from, because a failed machine can be quickly replaced with an identical one. When every minute is thousands of customers or employees unable to work, it’s worth the premium to be able to recover in seconds as opposed to hours.

Cloud: Costs, Costs Everywhere

Whenever somebody comes to us talking about going to an on-prem or collocated solution, it comes down to cost on the cloud side. Hardware is cheap these days and racking a Virtualization server that can run a dozen VM’s can cost less than a single middling SQL Server Instance on AWS or Azure. For organizations with dozens or hundreds of VMs, the cost difference can be an order of magnitude.

Worse, cloud costs tend to imperceptibly creep up over time. Slowly accumulating backup files without a limit, developers creating test machines and forgetting to delete them, decommissioned VM’s for systems that you’ve retired, but are afraid to outright delete. Bandwidth costs especially can be a nasty surprise. Whereas with an on-prem setup, you usually know from month to month exactly what you’re paying, with Cloud systems, the costs vary constantly, even ignoring pricing structure changes every few months.

The difference of several FTE’s worth of cost unquestionably drives many organizations to move back to physical datacenters, especially colocations where they share many of the capital costs with other companies.

Each organization needs to decide for itself what their risk and cost tolerances are, and almost all eventually settle on a mix of the two. For any given system, though, it’s important to keep in mind what the needs of the users are, and what the ability of the business is to support those needs.