1.8 Resource consumption

Drylab tracks your resource consumption across thre categories:

1. Tokens

What it is: The unit measuring how much AI processing your prompts and responses consume. Every message you send, every code cell generated, every database query interpreted costs tokens.

What consumes tokens:

• Sending a prompt to the AI assistant

• Generating or editing code cells

• AI planning and multi-step analysis

• Reading and parsing large files or notebooks

• Literature search and research queries

Tips to use tokens efficiently:

• Be specific and concise in prompts — vague prompts require more back-and-forth

• Avoid asking the AI to re-read large files unnecessarily

• Use Default Analysis mode for simple tasks (uses fewer tokens than Advanced)

• Break large analyses into focused steps rather than one giant prompt

2. Compute

What it is: The processing power consumed when running jobs — CPU cycles, GPU time, and memory usage during active computation.

What consumes compute:

• Running accelerated tools (protein folding, docking, alignment)

• Executing Nextflow pipelines (RNA-seq, variant calling)

• Heavy notebook computations (large matrix operations, ML training)

• Parallel jobs submitted to the cloud batch system

Compute is measured by:

• Job duration (how long a task runs)

• Instance type used (GPU costs more than CPU)

• Number of parallel jobs

Tips to reduce compute:

• Test on small data subsets before running full jobs

• Choose the smallest sufficient machine instance

• Cancel jobs that are clearly failing early

3. Storage

What it is: The disk space your files occupy in the Vault (/home/user/user_data/).

What consumes storage:

• Raw data files (FASTQ, BAM, H5AD)

• Pipeline outputs (alignment files, count matrices)

• Saved figures and result tables

• Notebooks and scripts

Tips to manage storage:

• Delete intermediate files after pipelines complete (e.g. raw BAM files after QC)

• Compress large files: .fastq.gz instead of .fastq

• Remove duplicate or unused datasets

• Keep only final results in the Vault; use temp workspace for intermediate work