How We Saved 28.1% On Cloud Compute
How Tracer identified and eliminated idle compute resources to save thousands
What we did
of runtime wasted on stale instances
of 348 running instances identified as stale
future monthly savings with no disruption to workloads
How We Saved 28.1% On Cloud Compute
In this case study
“We instantly identified 28.1% waste in our cloud infrastructure, unlocking $1,179.58 in recurring monthly savings with zero disruption”
Implementation
Overview
Tracer's internal AWS test revealed that 28.1% of runtime was wasted on idle servers and eliminating this unlocked $1,179.58 in recurring monthly savings while demonstrating the potential for six-figure annual savings at scale.
“With Tracer’s AI Cost Scan we unlocked a new level of efficiency and control by eliminating 28.1% of wasted cloud spend, proving the potential for optimization at scale.” — Vincent Hus, CEO and founding engineer at Tracer
Challenge
Here at Tracer, we run a mix of production, test and EC2 instances that are not always shut down after use, as in many organisations. Some of these remain active overnight, on weekends, or long after projects end. This avoidable inefficiency compounds quickly across environments into significant excess spend. These sleeping instances are burning money on servers that deliver no business value.
It is technically possible to track this type of issue. In AWS, you could use CloudWatch, but then you'd manually have to check every single instance individually, a process which simply is not scalable. Another option for AWS is to rely on AWS Trusted Advisor to take over this task, but this is a paid service and still requires ongoing effort to interpret and act upon.
On top of this, inefficiency isn't always as simple as idle servers. In high-compute environments, jobs can freeze as they get stuck in infinite loops, hung threads, or stalled processes. From the outside, these instances appear active, but in reality, they're delivering no results. In these cases, OS-level visibility is often the only way to recognize that jobs have stopped making progress.
With Tracer, this inefficiency becomes entirely avoidable. Our data shows that 28.1% of instance running time is stale and could be saved through early detection and automated termination turning wasted spend into recurring savings.
Tracer's instances view identifying idle compute resources across infrastructure, providing detailed analysis and optimization recommendations.
Internal Validation
This is how we did it internally. We deployed our own agentless monitoring system across our internal AWS accounts that had been active over a 90-days period, during which we ran a total of 348 instances. Within moments of deployment, the platform identified 101 out of 348 running instances as stale. By applying automated termination policies, we were able to stop wasteful spending and save over $1,179.58 in recurring monthly savings, with no disruption to active workloads.
Measurable impact
Most importantly, this internal test validates the savings potential at scale. When annualized, our internal account represents $41,783.85 in total runtime costs, of which $4,783.85 was overhead. When extrapolating to larger organisations with decentralized cost management and multiple AWS accounts, the impact grows dramatically.
While our internal validation focused on idle runtime, the same OS-level visibility can also surface frozen or stalled high-compute jobs. Catching these quickly could prevent massive overspend on large instances that appear busy but deliver no business value.
“It is really easy to use and in just minutes we had visibility across hundreds of instances and could act immediately, without touching our pipelines or disrupting workloads.”— Vaibhav Upreti, Founding Senior Engineer at Tracer
