Cloud vs. on-premises: Chasing the elusive cost comparison

Every business needs to consider cost when making the leap to the cloud, but there’s no benefit from apples-to-oranges comparisons.

calculator finance budget savings cost revenue
Irina Griskova / Getty Images

For most enterprises, “moving to the cloud” is not a mass migration. You have workloads running in your own data center. You have applications you want to add to the mix. On a case-by-case basis, whether migrating or building from scratch, should you opt for on-premises (on-prem) or the cloud?

Cost is only one factor in that decision – and in an era when the most exciting new technologies debut in the cloud first, probably not the most important one. Nonetheless, it’s a question most IT execs must at least attempt to answer when confronted by the C-suite or the board.

“I don’t want to move to the cloud and all of a sudden find out I’m paying double the amount of money – and have a big surprise for my CFO,” says Kim Weins, vice president of Cloud Strategy at Flexera, a software asset and cloud management tool vendor.

Hard-and-fast cost projections, though, are a different matter. If someone offers you a magic formula to determine whether running your workload will cost more in the cloud or on prem over time, they’re selling something. Given the dynamic nature of enterprise tech and the cloud in particular, think in terms of a rough estimate instead.

Managing cloud expectations

The basic pro-cloud argument has become almost cliché: Choose the public cloud and you don’t need to spend money on hardware or the people to maintain it. Instead of a capital expense, you’re talking an operational expense, and for new applications, quicker time-to-market than you could achieve on prem. The latter means you’ll see new productivity – or new revenue, if commerce is involved – sooner rather than later.

All of that is accurate enough, but it’s only part of the story. To begin with, as a million cloud newbies have learned, tossing workloads onto the cloud bears little relation to outsourcing.

“Sometimes people believe naively that if something goes to the cloud, it can take care of itself,” says Enzo Carrone, deputy CIO at SLAC National Accelerator Laboratory. “But that’s not true. And that’s a cold shower for lots of constituents and stakeholders.” That splash of reality is at its most bracing for IaaS, where engineers expert in navigating the complexities of AWS, Microsoft Azure or Google Cloud Platform tend to command big salaries.

Variables in the cloud-vs.-on-prem equation don’t end with the cost of personnel. Everything in IT is a moving target. On prem, that includes not only the inevitable obsolescence of physical infrastructure, but the fluctuating cost of AC power, data center square footage and licensing or subscription for infrastructure software. In the cloud, the cost of storage and VMs tends to fall over time – but exciting new services, such as those offering machine learning or serverless computing capabilities, can rack up enough pay-per-use charges to make a CFO blanche.

Searching for a data center baseline for cost comparisons

The typical data center and the workloads that run in it tend to become a massive tangle over time. Systems touch systems, applications share the same infrastructure, new things get bolted onto old things. Zane Whitner, vice president of product at Flexera, puts it this way: “From our experience – and this is across hundreds if not thousands of customers, all in the enterprise commercial space – most people don’t have a great handle on even what their current costs are.”

At SLAC, Carrone is at the beginning of his cloud journey, evaluating SaaS replacements for on-prem applications with an eye toward freeing up on-prem resources. “The calculation of the amortization and depreciation and moving from capex to opex is hard for us,” he says. Yet he is attracted to the benefits of SaaS, such as the capability to liberate his engineers from tasks like software patching and upgrading for on-prem applications. He also sees potential security and availability advantages.

In weighing the tradeoffs, Carrone has discovered that good-faith assistance from cloud providers can make a big difference. “The more they can educate us and help our IT staff understand these considerations in terms of pure financial management of these entities, the better we are positioned when have our financial analysts go down and do deep dives into what we’re prepared to do.”

In that vein, Whitner says that AWS recently bought TSO Logic, a product devoted to determining on-prem costs and helping customers save as they ascend to the cloud. “I think AWS was finding that this was a big roadblock for their deals,” he says. “For customers to go all-in, they needed more data, and I know [AWS] had several PhDs in economics building these massive Excel spreadsheets to help their customers do that analysis. But I think when the rubber meets the road, when you get to CIOs, that’s the data they wanted to see.”

But were they seeing reality? “If you think you can estimate down to the penny prior to migration what your bill will be,” Whitner says, “good luck.”

Cloud migration realities

Migrating an application to the cloud is a different kind of proposition than building a new one. For new applications, a wide distribution of users, the capability to scale on demand or attractive new cloud technologies may strongly suggest building on a cloud platform. However, if regulations dictate that the data used by an application must remain on-prem, that new app may need to stay in the data center regardless.

But when the application already exists and you decide to lift and shift it to the cloud, “if you just pick up and use the exact same resource allocation that you have on-prem, that’s probably overprovisioned and wasteful, even in a virtualized environment,” says Weins. “It’s always going to cost you more.”

Fortunately, few make that brain-dead mistake – or if they’re naive enough to do it, they quickly learn better. “You need to do at least a basic level of optimizing and right-sizing,” says Weins. “Are we actually using those resources that we’ve allocated to it? If not, then let’s right-size it as we move to the cloud. Does it really need to run 24/7? If not, let’s make sure we’re scheduling it as we move to the cloud.” She also recommends reserved instances, where customers can lower cloud costs by committing in advance to pay for resources and capacity, typically for one to three years.

If you do those basics,” Weins says, “then you’re probably going to save money right off the bat” by migrating a workload to the cloud.

Still, migration is often misdirected. If you’re simply moving existing applications built around flawed business processes to the cloud, you’d better off taking that same set of processes, cleaning them up, and building a new application using a modern cloud-native architecture with a bunch of cloud services that might help you in ways you hadn’t anticipated.

“It depends on the customer,” says Whitmer. “Some people may be in a situation where they’re like -- look, I’ve got a data center lease running out and I don’t want to have to renew it, so let’s just shove everything into the cloud as quickly as possible.”

Over time, though, you’re missing out if you don’t refactor those applications to take advantage cloud scalability, let alone cool new cloud services. And when you do, cost comparability with the original on-prem application goes out the window. “While It’s a little bit easier to compare apples to apples for infrastructure, compute, storage, network, for example, it becomes almost impossible when we get to some of these PaaS services, especially once you get beyond the most simplistic,” says Weins. 

Creeping cloud costs, boundless potential

The cost of those services can really add up. According to Weins, “as an IT industry, we’re not that great at understanding things like the cost of containers or serverless and what that looks like – or all of these other PaaS services that come into play.” High-level services that provide you additional value and agility can be pricey, she says, and the ultimate cost will be “super hard” to project.

Five to seven years ago, she sad, 80 percent of the cloud customer spend was on basic compute.

Today, for many customers, less than half goes to compute – and the rest goes to value-added cloud services. Yet Weins also noted that new cloud technologies can sometimes save money. Her own company reduced its cloud spend by a third by shifting to containers, “because we could now mix and match different microservices onto common boxes and common VMs and common instances.”

The bottom line is that the cloud is the right place for modern applications that need to scale and evolve to meet new demands. Rather than obsess over static cost comparisons, you’re better off monitoring your cloud workloads and cloud developers closely, largely to avoid paying for capacity or services you’ve stopped using. And when you want to take advantage of the latest edgy technology – be it machine learning, IoT connectivity, advanced analytics, or globally distributed relational databases – chances are it’s going to be ready and available on your cloud of choice.

As Whitner says, “if you’re moving to cloud purely on price, you’re doing it for the wrong reasons.”