Before you start, close unnecessary memory-heavy applications and check free storage. Raw reads, reference indexes, BAMs and temporary files can greatly exceed the final results folder.
1. Detect available resources#
On Compute resources, click Detect and Recommend. Review the proposed profile and resource values, then click Save Resources.
What resource settings change#
| Setting | What it changes | When to change it / example |
|---|---|---|
| Profile: balanced / low / high / custom | Chooses a resource allocation strategy. Changing the CPU or memory limit by hand switches the profile to custom, so the saved profile names the values the run uses. | Use a conservative profile when sharing the computer; custom requires a budget you can support. |
| CPU / threads | Limits computational parallelism. | More threads may shorten parallel steps, but do not multiply RAM capacity or speed every step equally. |
| Memory budget | Constrains local job scheduling and must fit available RAM. | Leave room for the operating system and other processes. Assigning 32 GB on a 16 GB machine does not supply memory. |
| WSL2 memory / CPU limits | Caps Linux resources on Windows. | Raise a restrictive cap only within host capacity and apply the required WSL restart safely. |
| Aligner selection | Changes memory/storage needs and computational route. | STAR is memory-intensive; HISAT2 and Salmon offer alternative routes. Salmon does not produce BAMs. |
2. Inspect the runtime estimate#
Open Runtime estimate and click Estimate Runtime after saving inputs, reference and resources. Use it to plan, not as a completion deadline.
If a job is killed or storage fills#
Read the rule log to distinguish out-of-memory, disk exhaustion and tool errors. Reduce concurrency or choose a justified alternative route only after identifying the cause. Do not repeatedly resume while the same resource shortage remains.
Advanced: cluster allocations
Local CPU/RAM pools are not the same as a scheduler’s per-rule allocation. The published Slurm and Kubernetes profiles avoid a global local-memory pool. Review the generated command and site-specific resources before submission; see HPC guidance.