Quick Win #3

And a Few More Great Wins!

Quick Win #3

Stop Cloud Overprovisioning

According to Amir Sharif, vice president of product and strategy at Opsani, right-sizing your existing cloud workloads is one key to cost savings. “Nearly all of your cloud workloads are overprovisioned,” he claims, while noting a still deeper issue:

Developers focus on getting the software to work, not getting it to work at the right cost. They have an invisible incentive to oversize. That invisible incentive is driving up your cloud costs.

Amir Sharif, VP of Product and Strategy at Opsani

You often hear about cost/performance trade-offs in IT. I hadn’t thought incentives behind DevOps and Agile might drive more toward performance, less toward cost-effective architecture. Shouldn’t principles 9 and 12 in the Agile Manifesto apply to technical “cost” excellence and becoming more “cost”-effective, too? Densify's Chuck Tatham says:

IT organizations should stop allowing developers and other technical staff to choose cloud resources or instance types manually through educated guesses or assumptions. Engineering and development staff are oriented toward ensuring performance and stability, and not cost efficiency. So, they naturally over-specify resources. At scale, this is hugely wasteful. There are automated ways to make selections empirically.

Densify’s Chuck Tatham

More Wins

Take Advantage of Cloud Purchase Discounts

More Wins

Showback/Chargeback and Spot Instances

We’ve avoided going into other, more obvious wins so far. But it’s important to touch on a few more. One word of advice: using many of these can yield significant cloud cost savings, but they may require more careful planning and implementation.

First up for other, potential wins is the use of cloud purchase discounts. Here, we mean things like reserved instances, reservations, or Savings Plans (in the world of AWS). These rely on an organization’s longer-term commitment to buy and use a certain amount of cloud resources. Reservations provide significant savings over pay-as-you-go prices.

“Purchase reservation discounts” sounds like simple and easy advice, but in practice, it’s quite challenging for organizations to buy reservations effectively at scale. Here’s some guidance on the use of cloud purchase discounts:

Consider starting small. Making the most of these types of volume discounts can take some time to figure out. Starting with small, frequent reservation purchases may allow your organization to learn as you grow.
Don’t miss the forest for the trees. According to SHI's Communications, Media, and Entertainment CTO, Lee Ziliak, use of things like reserved instances or spot instances can seem like quick wins from a purchasing standpoint, but organizations should also look more strategically at the bigger picture. This includes evaluating other, longer-term discount vehicles for potential cost optimization, such as programs like the Amazon Web Services Enterprise Discount Program (AWS EDP) or the Microsoft Enterprise Agreement (EA).
Says Ziliak, “Such other, longer-term discount vehicles should be included in a strategic approach that requires more planning and time to negotiate.”
Review the fine print. Though not just about cloud purchase discounts per se, organizations should also take time to revisit their current vendor licensing terms, payment terms, and contracts. Often, there’s a fair amount of wiggle room here for potential cost savings. In Stratascale’s experience, many organizations miss several, little-known ways to save quite a lot of money in these areas.

Take Advantage of Cloud Purchase Discounts

What is a Spot Instance?

Spot instances allow organizations to bid on excess capacity, provisioning virtual machines on-demand at a rate up to the price they bid.

The going market rate for excess compute is known as the “spot rate,” and can offer up to a 90% discount on the on-demand rate.

Spot instances may be terminated at a minute’s notice, if the market rate rises above the customer’s bid price, or the cloud provider runs out of excess capacity.

To spot (or not) to spot. We made brief mention of spot instances earlier, but the jury is actually mixed about whether spot instances are a good idea.

On the one hand, Opsani’s Sharif isn’t a fan of this type of “cloud-real estate arbitrage,” even if you’re planning to use it for an “ephemeral or experimental workload,” he says. “This type of back-and-forth to shave a few dollars here and there is not worth the hassle and the additional operational costs."

On the other hand, a Red Hat Continuous Kernel Integration (CKI) project team made a pretty good case for the use of spot instances to help cut their own continuous integration costs for kernel testing. Granted, getting to cost savings required a few extra learning cycles by the team.

The truth is probably somewhere in the middle: spot instances may work for certain workloads or use cases, especially for workloads flexible enough to handle disruption or being run at any time.

Implement early showback or chargeback. When implemented correctly, both showback and chargeback let departments know how much IT resources they are using. According to a BMC blog on the subject, showback and chargeback are closely related. Showback makes departments aware of how they are using IT resources by showing them the cost. Chargeback goes further and holds departments accountable for spending by actually charging them for usage.

Implementing showback or chargeback for the cloud is often part of broader organizational efforts toward cloud cost optimization. Still, SHI’s Ziliak notes it can be fairly easy and quick to get a head-start in this area. Doing so also offers good incentive for people and organizations to behave better with cloud spending. (In our recommendations section, we share more on both the ideas of starting small and incentivizing good cost-optimization behaviors.)