Focus on your applications with Continuous Optimization
Paid feature The cloud is the future. The problem for most organizations is that when it comes to working out what resources they need for that future – and what they’re going to cost – they are usually flying blind. As organizations move more and more complex workloads to the cloud, they need to be sure they can allocate resources and costs efficiently, and pinpoint where there is potential waste and overspend. The problem is, at best, the tooling provided by the main cloud platforms are basic visibility tools that show where they’ve been and give the merest hint of what things will look like in the very immediate future. This was the situation facing Spot’s founder, Amiram Shachar when he was working as a director of DevOps while studying computer science. « I had a real problem… our cost of cloud was way off what the calculators we used to predict and project how much money we were going to pay. » Cloud vendors’ tools would allow him to visualize his bills, but this was all retrospective. Back then, the only way to get a grip on the problem was to take time out with a stack of price lists, and a spreadsheet. And this all ran counter to the continuous improvement philosophy behind DevOps. « I wanted to make it a continuous thing that I’m always optimizing costs, » he says. From there, the obvious step was to recognize that « the machine can do it better than me. » So, as part of his studies, he prepared a project about optimizing cloud costs through sophisticated pricing selection of compute instances. That project became Spot, which aims to help businesses make the most efficient use of the multitude of services and pricing options available in the cloud. The company was snapped up by NetApp in 2020, and Shachar remains its vice president and general manager. The Spot by NetApp model is straightforward. Cloud vendors offer three types of resource. Shachar describes standard pay by the hour on-demand pricing as « very linear and the most expensive way to buy cloud ». Reserved instances will attract a discount in the tens of percent – but only if a user commits for a period of time. The problem is, « you need to do a lot of planning work. » The biggest potential discounts come on spot instances, the spare capacity that cloud platforms have available at any given moment. These can be as much as 90 percent cheaper. That’s compelling, if you can live with the fact that the cloud vendor they’ll kick you off with zero notice if it needs that capacity. This poses an obvious operational challenge, and consequently the option was historically unfit when it came to serious work. It might seem that cloud providers would have no interest in helping customers use the cheapest possible instances. But says Shachar, this ignores the Jevons paradox that « the more optimized, the more efficient, you become in consuming a resource, the more of that resource you consume. » Also, « cloud providers saw that unoptimized customers tend to churn much faster than optimized customers…they know if they optimize they’ll spend more, they’ll use more, and they’ll stay longer.
