We Saved Up To 39.7% On Our AWS Batch Bill

Bioinformatics teams rely on AWS Batch to run their workflows such as RNA-seq. However, running these workloads across multiple instance families and regions often leads to unpredictable and inflated costs as AWS Batch automatically selects which instances to use.

Oversized instances burn unnecessary resources, region selection impacts runtime and cost efficiency and researchers spending time on experimenting configurations instead of science.

Therefore, organizations need a reliable, data-driven way to control AWS Batch spending without costly or risky pipeline rewrites.

At the same time, inefficiency is not limited to oversized instances. In high-compute workflows, pipelines can get stuck in infinite loops, hung threads or stalled processes. These frozen jobs keep burning compute resources and from the outside they appear active even though no progress is being made. Detecting these requires OS-level visibility to identify when jobs have stopped advancing, preventing large amounts of wasted spend.

At Tracer, we ran controlled RNA-seq workloads on AWS Batch starting from the default AWS Batch configuration. We monitored the most significant cost and performance factors to understand where optimization was most beneficial.

We examined thoroughly on-demand instance families (c5, m5, r5, etc.) and performed numerous cost experiments to quantify where real savings could be made, to identify the best price/performance balance and validate the cost reductions without any pipeline rewrites.

In order to do this testing in a controlled environment, we configured compute environments to restrict allowed instance families so each run isolated one family at a time.

Furthermore, we ran controlled benchmarks across different instance strategies and included multi-region testing to show regional cost differences. We used our centralized observability platform to track costs and performance.