Run Nextflow on AWS Without Changing Your Pipeline
Fovus benchmarks every process and auto-assigns the right AWS instance, storage, and Spot strategy — swap one config file, keep everything else.
The challenges
Nextflow pipelines like nf-core/rnaseq and nf-core/sarek pack memory-bound alignment, single-threaded QC, I/O-saturating BAM conversion, and embarrassingly parallel variant calling into the same run — yet typically execute under one human-defined resource configuration shared across every step. Lightweight processes get overprovisioned, demanding processes fail and restart, and most of the AWS EC2 instance catalog goes unused. As input sizes grow, the config that worked last run stops fitting the next one, and pipelines fail with out-of-memory errors and stalled jobs rather than scaling cleanly.
How we help
Fovus auto-benchmarks every Nextflow process individually against representative inputs — profiling bottlenecks, resource consumption, and the impact of applicable HPC strategies, free of charge and with no user effort.
Fovus feeds those benchmark profiles into an AI optimization engine that assigns each process its own EC2 instance type, storage configuration, parallel computing settings, and Spot/On-Demand routing — automatically, at runtime.
Fovus automatically updates benchmarking profiles as AWS rolls out new instance families, so your pipelines stay on the optimal configuration — no workflow changes, no manual re-tuning.
Fovus snapshots process state at intervals and on Spot interruption, so runs auto-resume from the last checkpoint with Spot-to-Spot failover — making Spot practical for long-running processes.
Fovus provisions a POSIX-compatible distributed file system using local SSD as a transparent Amazon S3 cache, sized per-process based on measured I/O demand — built for I/O-intensive steps like BAM conversion.
Fovus applies the guardrails you set — configurable constraints on optimization scope, AWS Regions, and spend — enforced automatically at the infrastructure layer.
The Numbers
Not the Marketing
Real benchmark data from nf-core/rnaseq and nf-core/sarek on AWS — no synthetic workloads, no rounding in our favor.
What it costs
$0.70
Per sample
nf-core/rnaseq cost
Actual AWS cost per sample on Spot with memory checkpointing — test_full, 8 samples, ~100M paired-end reads each.
$6.15
Per 30× WGS
nf-core/sarek cost
Full 30× germline WGS — SNV/indel, SV, CNV, annotation — on Spot with memory checkpointing, CPU-only.
What you save
70–85
%
Cost avoided
Typical overspend eliminated vs. one static resource configuration applied uniformly across all pipeline steps.
3–7
X
Dollar efficiency
Higher dollar efficiency vs. single-config cloud deployments, benchmarked on Spot with memory checkpointing.
Business Impact
Genomics teams running nf-core/rnaseq and nf-core/sarek on AWS with Fovus are seeing real results: ~$0.70 per sample for RNA-seq and $6.15 for a full 30× germline WGS run, both on Spot with memory checkpointing — 70–85% lower cost than a single static configuration.
Your pipeline could be next.
1/1