In the environment of on-premises data centers, it was straightforward enough: keep the servers running 24/7. This attitude, founded on huge capital investments and necessity to maximize hardware uptime, has carried into the cloud in dangerous fashion. The outcome is the silent killer of the budget otherwise referred to as the always-on idle cloud resource that stretches across organizations of all sizes.
Although the pay-as-you-go model of the cloud is most flexible, our habits of using things often betray the main economic merit of the cloud. We put development servers and testing environments, staging instances and even keep them on full time- all night, all weekends and all public holidays. This virtual ghost town of idle compute resources costs a great deal of money.
The problem of this magnitude has been consistently noted in industry reports. The Flexera 2023 State of the Cloud Report reflects a current trend toward cloud cost optimization, with optimizing existing cloud spend being ranked as the number one issue by organizations, and by themselves when the same organizations self-reported their estimated wastage on a predetermined range of 30-32 percent. A large percentage of this waste can be directly attributed to idle resources -compute instances that are provisioned and active but are not doing any useful work.
This is a not a minor leak; a gaping hole in the hull of your IT budget.The Math of Waste: Why You Pay for Idle
The Math of Waste: Why You Pay for Idle
The costing mechanisms of the leading cloud providers include AWS, Microsoft Azure, and Google Cloud Platform (GCP), but it relies on allotted resources, rather than on consumed resources. When you spin up an AWS EC2 instance, an Azure Virtual Machine, or a GCP Compute Engine instance you are paying when its running, not just when its CPU is fully in use, at 90 percent or at 0.1 percent.
Consider a typical nonproduction environment, perhaps used on a development and testing basis. It is mostly used during normal working hours of around 8 hours per day and 5 days per week. This is an equivalent of 40 hours of active use a week. But there are 168(24hour x 7 day) hours in a week.
Assuming the latter environment is run 24/7, you are paying 168 hours of uptime but only getting value out of 40. Redundancy at such a level implies that there are more than three out of four wasted expense dollars spent on such resource.
The Waste Percentage = Total Hours 168 (168 40) =76.2 percent
Multiply this across dozens or hundreds of non-production instantiations, and the cost implication is huge. It is like leaving all the lights switched-on inside an empty office building weekend after weekend after weekend.
The Solution: Intelligent Scheduling and Automation
Automated scheduling is the best strategy to combat this silent leakage since it is difficult to turn resources off manually (due to human error) and is, also, cumbersome to scale. This technique is commonly called Cloud parking, start/stop automation. This entails configuring policies that will terminate resources on demand and resume under demand.
Such a strategy brings cloud expenditure into direct correlation with real-life demand maximizing the cloud elasticity inherent.
The best bets to schedule
The most strategic place to begin is non-production environments that often make up a significant amount of an organizations cloud footprint. Prominent candidates will be those who reveal the importance of finding a solution.
- Development and Test Environments: These are usually normally only used during work hours.
- QA Environments and Staging: These types of environments can then be off between testing cycles.
- Virtual Desktop Infrastructure (VDI): Only the time employees are working they need desktops.
- Training and Demo Platforms: Refers to platforms used occasionally (and can be switched on-demand).
- Batch Processing and Analytics Jobs: Processes that execute in predictable, well known time windows.
Production environments are typically not included, but components with cyclical demand levels (such as multiple tiers of applications) may optionally be scaled back, up to and including turn-off.
Tools for the Job: Implementing a Scheduling Strategy
An effective scheduling strategy can now be more easily achieved because there is a variety of the tools that are offered by cloud providers and are provided by third-party specialists.
1. The Native Cloud Vendor Tools
Each of the major cloud providers provides solutions to connect to automate out of the box:
- AWS: AWS Instance Scheduler is a pre-packaged solution that runs on Lambda functions and CloudWatch used to initiate turning on/ off EC2 and RDS instances as per the custom schedule.
- Microsoft Azure: Azure Automation enables to create runbooks, which are powerful scripts to automate the start-up/shut-down process of VMs on some schedule. A simple, but often effective starting point is to use the Start/Stop VMs during off-hours feature.
- Google Cloud Platform: GCP provides Instance Schedules to its Compute Engine, which lets you add schedules to individual or any number of VM instances so that you can preset their availability to operate automatically.
These native tools are very powerful, cost effective and an excellent means of demonstrating the worth of scheduling to your organization.
2. Third Party Cloud Management
With more advanced functionality, third-party platforms are useful to organizations with multi-cloud setups or those that want more robust feature sets. Such tools as Harness Cloud Cost Management (previously ParkMyCloud), CloudHealth, or Flexera One offer:
- Managing multi-cloud in a centralized manner through a single dashboard.
- Policy or rules-based automation, which may be by tag, resource groups, or other metadata.
- A snooze option which allows developers to exceptionally override schedules.
- Savings reports that show your use of money in scheduling policies visually.
Overcoming the Hurdles
The advantages are obvious, although the obstacles on the way to their implementation also exist. Typical challenges include stateful applications which necessitate an orderly shutdown, complexity of inter-dependencies between resources, and cultural resistance due to teams which are not used to an always-on policy.
The trick is to get a small bang. Make the first development environment the least risky. Hammer the idea home with the group, set forth a clear timetable, and show them the savings. Use this success to establish momentum and to establish best practices in scripting an orderly shutdown and startup of dependent services.
It is a legacy on-premises concept to leave cloud resources running all the time. It is a misconception of the value promise of the cloud. The cloud does not simply mean a server in someone elses hands; it is an adaptive and flexible system that aims at efficiency.
When organizations implement intelligent scheduling, they can end the passive drain of tens of thousands–or even millions–of dollars spent on idle compute. Check your inactive systems through auditing. Determine the “digital ghost towns” that are going through nights and weekends. By adding a simple start/stop schedule you can instantly turn that waste into savings, liberating funds to invest in innovations instead of inefficiency.
