How to avoid wasteful cloud expenses

Companies of every size, from the enterprise giants to their SME counterparts need to find ways to manage the cost of cloud services.

Gartner forecasts end-user spending on public cloud services to grow 21.7% to reach $482 billion this year.

According to Aran Khanna, CEO Archera, a cloud cost and resource optimisation company, a third of cloud spend is wasted through overcommitting or getting locked into the wrong one and three-year contracts.

As a former AWS engineer, Khanna says one reason for this, is that it is the nature of commitments. “In exchange for making a commitment to spending a certain amount on resources for the next one or three years, companies are given a significant discount.”

And the less flexible a contract is, the higher the discount, he says. “Contracts that are shorter term or can be exchanged trade off savings for added flexibility. For companies looking to save money, these discounts are quite significant and very enticing. But, the data and tools used to select commitments are limited in their ability to forecast usage.”

He says forecasts offered by most tools only look at historical data to predict future needs and recommend one- or three-year contracts. The problem is that historical usage is not always reflective of how the organisation plans to move forward.

“Events such as the introduction of a new product, the impact of migration and right-sizing, or even increased sales can cause a drastic deviation from historical trends. Even if you know your business is going to be making these changes, commitment recommendations will not reflect that. If the impact of these events is not taken into account, you are at a risk of selecting resources that may not be appropriate for your infrastructure in the future but cannot change due to lock-in with the commitment.”

So what is the solution? Khanna says it's scenario planning tools that allow users to model different outcomes and how they would impact the spend. 

“Establishing a higher and lower bound for spend helps identify how much infrastructure can safely be covered by high savings but low flexibility commitments, and what should be covered by higher flexibility but lower savings commitments.”

Also, he says guaranteed buybacks for standard reserved instances (RIs) should be considered. “If there was a guarantee that your unused standard RIs would be sold in the AWS Marketplace, there is a lot less risk in this commitment. You can maximise the discount provided by standard RIs but can sell it back before your term ends if your needs change. The option of a guaranteed buyback can accomplish this.”

The reason there’s a tradeoff between flexibility and savings, says Khanna, goes back to the issue of how commitment discounts are structured. Three-year standard RIs offer the most savings relative to on-demand pricing, but you cannot get rid of it if you don’t need it. A one-year standard RI is more flexible because it is a shorter commitment, but it does not provide as significant savings.

Convertible reserved instances (CRIs) are significantly more flexible than standard ones, because of their exchangeability. “The result is that even with a longer three-year CRI commitment, you will only get the savings of a one-year standard RI.”

He says AWS introduced savings plans to deal with this issue. “Instead of committing to using a certain number of instances like reserved instances, you commit to spending a certain amount of money. For many companies they are easier to use because the discount is automatically applied to new infrastructure within the same tenancy and OS, and in any region. You have to do this manually for reserved instances, and once again there is a tradeoff, as savings plans save teams less money than RIs.”

There are three ways to pay for each contract: all upfront spend, partial upfront spend, and no upfront spend. “All upfront spend can immediately use up a team’s budget, but it offers the most savings," says Khanna.

According to him, to find the right balance of flexibility and savings, the types of contracts being used must be blended. “You need to consider every variation of contract type (RI, savings plans, on-demand), every term length (one, three or five years), and upfront spends to find the purchasing strategy that meets your organisation’s financial and infrastructure needs.”

He says that’s about 36 different options for each instance, so this is an onerous task if done manually.

Unfortunately, most tools offering purchasing recommendations require the user to specify a contract type, term length, and upfront spend which is then applied to all commitments, meaning every option is not being scrutinised and the user is stuck taking a one-size-fits-all approach.

For Khanna, the solution is using a tool that does real optimisation across every potential commitment domain for the entire cloud resource portfolio. “This way you will not only find commitments that are appropriate for individual resources in your infrastructure, but also work with every other commitment to balance the flexibility and savings of all your resources.”