9: Process Management, Multi-Core Processing & GNU Parallel

Chapter 9: Process Management, Multi-Core Processing & GNU Parallel

Genomic analyses are compute- and memory-intensive: aligners saturate 64 CPU cores, and de novo assemblers can consume hundreds of gigabytes of RAM. Managing system resources, keeping long-running pipelines active across network disconnections, and parallelizing repetitive sample processing across all available CPU threads are critical competencies for any bioinformatician.

1. Inspecting Running Processes & RAM

# Interactive real-time process monitoring (CPU and memory consumption)
top
htop  # Enhanced visual process monitor (install via conda/apt)

# Check available system RAM in human-readable gigabytes
free -h

# Check available hard drive disk space on all mounted filesystems
df -h

# Find specific running processes matching 'bwa'
ps aux | grep bwa

2. Background Execution & Job Control

# Append an ampersand (&) to launch a process directly into the background:
fastqc sample.fastq.gz &

# Job control shortcuts:
Ctrl + C   # Terminate (interrupt) current foreground process
Ctrl + Z   # Pause (suspend) current foreground process

# View all active background jobs in the current shell session:
jobs -l

# Resume a suspended job in the background:
bg %1

# Bring a background job back into the foreground:
fg %1

3. Persistent Sessions with tmux

tmux (Terminal Multiplexer) creates persistent sessions that survive network drops, computer sleep, and SSH timeouts:

# Start a new named tmux session
tmux new -s rnaseq_run

# Inside tmux: Launch your pipeline normally
./run_full_rnaseq_cohort.sh

# Detach from the session safely: Press Ctrl+b, then release and press d
# Your pipeline continues running in the background on the server!

# Later (from any computer): Re-attach to your running session
tmux attach -t rnaseq_run

# List all active tmux sessions on the machine
tmux ls

4. Terminating Unresponsive Processes

# Send a graceful termination signal (SIGTERM 15) by Process ID (PID)
kill 12345

# Send a forceful, unignorable kill signal (SIGKILL 9) if the process is hanging
kill -9 12345

# Terminate all processes matching a specific tool name
killall bwa-mem2
pkill -u $USER samtools

5. High-Throughput Scaling with GNU Parallel

Instead of processing samples sequentially in a slow for loop, GNU Parallel distributes tasks across all CPU cores simultaneously:

# Run FastQC on all FASTQ files using 16 concurrent CPU jobs:
parallel -j 16 "fastqc {} -o qc_reports/" ::: *.fastq.gz

# Parallelize paired-end read alignment across multiple samples:
# Given a list of sample names in samples.txt:
parallel -j 4 --colsep ' ' \
    "bwa-mem2 mem -t 8 /ref/GRCh38.fa {1}_R1.fq.gz {1}_R2.fq.gz | samtools sort -o {1}.bam -" \
    :::: samples.txt