Reining in Kubernetes Costs in the Cloud
IT organizations have been overprovisioning infrastructure since the dawn of IT, but a survey published today by StormForge, a provider of performance testing tools, suggests the addition of Kubernetes clusters to IT portfolios may be further exacerbating the issue.
The main reason IT infrastructure is overprovisioned is because developers and IT teams want to make sure each application has enough resources to maintain availability. The trouble is, ensuring availability by overprovisioning comes at a cost.
The StormForge survey of 105 IT professionals in North America finds that nearly half of cloud resources spending (48%) is wasted. Reducing that waste is, to varying degrees, a priority for three-quarters of organizations (76%), with a third (33%) citing it as a high priority, the survey finds.
That focus on waste correlates with organizations’ plans to increase cloud spending over the next 12 months, with just under a third of respondents noting they will significantly increase cloud spend over the next 12 months.
Jacques Penicaud, vice president of marketing for StormForge, notes much of that increased spending is going to be driven by deployment of cloud-native applications on Kubernetes clusters.
The survey finds roughly three-quarters of respondents are already running Kubernetes clusters, with just under half running 11 or more. A third of those respondents (33%) said they rely on the default settings from cloud service providers to size their clusters, while just under a third (31%) rely on a trial-and-error approach. Surprisingly, just over a quarter (27%) said they are already relying on machine learning algorithms to properly size Kubernetes clusters.
Kubernetes, of course, is one of the most complex platforms IT organizations manage today. The survey finds that complexity is either a major reason Kubernetes clusters are overprovisioned (27%) or that it is at least a contributing factor (36%).
Penicaud says that as adoption of automation frameworks and testing tools based on machine learning algorithms become more widely adopted, the level of overprovisioning in cloud computing environments should decline as Kubernetes becomes easier to deploy and manage.
Interestingly enough, a full 94% of respondents claim the predictability of monthly cloud spending is not an issue, with 94% saying they know, at least roughly, what their cloud spend will be each month. The issue is gaining more control over that cloud spending, notes Penicaud.
In the wake of the economic downturn brought on by the COVID-19 pandemic, IT organizations are under increased pressure to reduce costs. The challenge they face is that many cloud resources are provisioned by developers that don’t always have the greatest appreciation for costs. A developer’s primary concern is to make they don’t get a support call informing them the performance of an application is degrading because there are not enough compute and storage resources available. IT operations team that are tasked with paying the bill for cloud services, however, tend to be extremely concerned about total costs.
It may be a while before those IT operations teams can more aggressively rein in those cloud costs. However, it’s now just a matter of time before an increased focus on cloud optimization yields a better financial result.
