Firms' cloud server capacity is only being half-used as they see over-provisioning as a "necessary evil", according to research from cloud provider ElasticHosts.
Cloud server provider ElasticHosts questioned 200 CIOs for its study and found that most cloud customers are "over-paying" for cloud servers, while still failing to get the peak performance they need.
Yet despite this, almost half believe this is still "delivering on the promise of cloud". Traditionally, cloud servers have been paid for according to the levels of capacity provisioned - if you run a 4GB server then you pay for a 4GB server, regardless of whether that capacity is being used.
The study found that businesses use just half (51 percent) of the capacity they provision over a 24/7 cycle. As a result, customers are effectively paying twice as much as they should be for the capacity they are using, says ElasticHosts.
When looking at the reasons behind this over-provisioning, 90 percent of respondents said they see over-provisioning as a necessary evil in order to protect performance and ensure they can handle sudden spikes in demand.
However, despite such over-provisioning, performance is still suffering. Many companies are unwilling to pay extra for even more capacity that they only use a small amount of the time, with 88 percent of CIOs admitting they often "sacrifice" peak performance in order to keep costs down.
"This research really highlights the short-comings of the pay-by-capacity billing model," said Richard Davies, CEO of ElasticHosts. "Essentially, companies are paying for space they are not using half the time because they are running with extra headroom so they can handle peaks in performance.
"Yet at times of really high demand, when arguably it is even more important that everything is working smoothly, web applications or websites will run slowly or even fail because they do not have the amount of server capacity needed."
Davies said changes are afoot however, as updates to the Linux kernel have enabled a new generation of container technology, providing an automatically scaling infrastructure that can be charged by usage, rather than by upper capacity levels. Davies said traditional virtual machine servers do allow users to scale their servers up and down according to need.
Although companies can create APIs that automate this process, many don't have the expertise to do it themselves. They also don't have the additional resources to enable a systems administrator to sit and monitor capacity consumption all day, meaning levels are often just set and left to run.
This means many cloud services being supplied do not expand and retract automatically to meet demand, and are not being paid for like a utility - elastic usage and a utility-based charging model based on actual consumption were two key hype factors around the cloud.