Organizations that migrate existing on-premises workloads to the cloud without revisiting the way services are provisioned are throwing away money, according to a new study.

Cloud migration and optimization vendor TSO Logic Inc. looked at the usage patterns of 105,000 operating system instances from 20 customers and prospects over a six-month period and found that 84 percent were overprovisioned, meaning their organizations had allocated more processor power or memory than the workloads required.

If those workloads were moved directly to the cloud without modification, costs would increase nearly 20 percent, the company estimated. However, if cloud resources were adjusted to match the needs of the workloads, costs would fall by over 35 percent.

TSO Logic has an obvious interest in these findings because the company’s business is helping clients migrate cost-effectively to the cloud. Still, the results are interesting look into the risks of a “lift and shift” cloud migration strategy.

Find the right fit

Cloud providers collectively offer hundreds of different pricing plans, giving their customers enormous flexibility to pick and choose exactly the resources they need. When moving an existing workload, organizations may instinctively make what they believe is the safe choice and choose the same combination of processor and memory resources in the cloud that they’re running onsite. However, if onsite utilization is only 20 percent, such a strategy results in drastic overprovisioning and wasted cost.

“Overprovisioning happens because companies don’t know how much compute they need or because they have limited choices,” said Aaron Rallo, chief executive of TSO Logic. “We found a lot of Ferraris that were being driven like Volkswagens – expensive technology being used very little.”

Provisioning is as much art as science, particularly in a variable workload scenario. In some cases, overprovisioning is intentional. Retailers, for example, may let large parts of their computing infrastructure go unused for much of the year so that they have enough power in reserve for the holiday season. However, a lot of overprovisioning is also the result of bad planning or just negligence.

TSO Logic’s research belies the assumption that cloud infrastructure is inherently cheaper than on-premises computing. On an apples-to-apples basis, the cloud is actually more expensive. The cost leverage comes from using elastic instances that scale according to demand and shutting down instances when they’re not being used.

In collecting data for its report, TSO Logic gathered hundreds of millions of data sets across more than 15 utilization metrics, Rallo said. It found that workload characteristics vary widely. In some cases, utilization may run less than 10 percent for 29 days out of the month, but jump to 100 percent on day 30. Some companies also don’t pay particularly close attention to utilization because they have had limited choice when choosing on-premises resources.

Evaluating resource utilization isn’t a point-in-time thing, since many workloads vary according to factors such as seasonality. “Organizations need to understand how they are truly utilizing each workload to accurately evaluate cloud options,” the report said. “The more data collected over time, the more accurate those evaluations become.”

Most cloud providers offer combinations of elastic and fixed instances, with discounts for customers that are willing to make multiyear commitments to a set configuration. For workloads with predictable resource requirements, those are often the best deal, Rallo said.

Elastic pricing is more expensive but may be cheaper in the long run by offering the ability to flex up and down. “If you’re using the platform three hours a day, it’s a better candidate for on-demand than if you’re using it 24 hours a day,” he said.

The bottom line is that there is no one perfect infrastructure-as-a-service solution, but there are a whole lot more options than customers have ever seen before.

Image: Flickr