Conda Performance Optimization Guide

Critical Performance Issues

Root Cause: Solver Algorithm

• Legacy conda solver: Uses PycoSAT (Python-based) for dependency resolution
• Performance impact: 40+ minute install times for basic packages like jupyter
• Failure mode: Downloads 100+ MB metadata, builds complex dependency graphs, tries all combinations
• Breaking point: Often fails completely after 45+ minutes on environment updates

Solution: libmamba Solver

• Available since: conda 23.10.0
• Technology: C++ implementation using libsolv (same as Linux distros)
• Performance improvement: 43 minutes → 90 seconds (90%+ reduction)
• Critical threshold: 60% of users abandoned conda for mamba due to speed issues

Configuration Requirements

Solver Configuration

# Check current solver
conda config --show solver

# Should return "libmamba" - if "classic", update immediately
conda install -n base conda-libmamba-solver
conda config --set solver libmamba

Channel Optimization

# Remove slow default channels
conda config --remove channels defaults
conda config --add channels conda-forge
conda config --set channel_priority strict

• Critical: strict mode prevents 12+ minute version comparisons
• Impact: Picks first matching package instead of comparing all channels

Cache Management

# Check cache size (often 12+ GB)
conda info

# Clean operations
conda clean --packages      # Safe: removes old downloads
conda clean --index-cache   # When solver gets stuck
conda clean --all          # Nuclear option for broken state

Performance Comparison Matrix

Tool	Typical Install Time	Use Case	Critical Notes
Legacy conda	18-43 minutes	Avoid	6x slower than pip
Modern conda (libmamba)	3-4 minutes	General use	2x slower than pip
Mamba	90 seconds	Speed-critical	Drop-in replacement
Micromamba	30 seconds	Docker/CI	5MB vs 500MB bloat

Critical Failure Scenarios

"Solving Environment" Hangs

• Timeout threshold: If running >10 minutes, it's stuck (not thinking)
• Solution: Ctrl+C, then conda clean --index-cache
• Root cause: Corrupted metadata or impossible dependency constraints

Storage Exhaustion

• Cache growth: 12-50GB common for old installations
• Risk: Hard links from pkgs/ folder to environments
• WARNING: Never manually delete pkgs/ folder - breaks all environments
• Safe cleanup: Only use conda clean commands

Package Installation Patterns

# WRONG: Triggers multiple solves
conda install numpy
conda install pandas
conda install matplotlib

# CORRECT: Single solve operation
conda install numpy pandas matplotlib scipy scikit-learn

Migration Strategies

Immediate Performance Gain

1. Solver upgrade (if conda 23.10+): conda config --set solver libmamba
2. Channel optimization: Switch to conda-forge only
3. Cache cleanup: conda clean --all
4. Expected improvement: 80-90% time reduction

Alternative Tools Installation

# Mamba (full compatibility)
conda install mamba -c conda-forge
# Use: mamba install [packages]

# Micromamba (minimal footprint)
curl -L micro.mamba.pm/install.sh | bash
# Use: micromamba create -n env python numpy pandas -c conda-forge

CI/Production Recommendations

• Avoid conda in CI: Use micromamba or pre-built Docker images
• Performance gain: 15 minutes → 2 minutes build times
• Use environment.yml: Faster than iterative installs
• Cache strategy: Cache package directory between builds

Common Misconceptions

"Conda is Always Slow"

• Reality: Only legacy solver (pre-23.10) is fundamentally slow
• Modern performance: Acceptable for most use cases with proper configuration

"Multiple Conda Commands Can Run Concurrently"

• Reality: Conda uses file locking - only one command at a time
• Failure mode: Corruption if attempted

"Pip and Conda Don't Mix"

• Best practice: Install conda packages first, then pip packages
• Avoid: Installing conda packages after pip in same environment

Troubleshooting Decision Tree

Install Takes >5 Minutes

1. Check solver: conda config --show solver
2. If "classic" → upgrade solver
3. If "libmamba" → check channels, clean cache
4. Still slow → switch to mamba

"Package Not Found" Errors

1. Add conda-forge: conda install -c conda-forge [package]
2. Clean index: conda clean --index-cache
3. Check package exists: Search on anaconda.org

Environment Corruption

• Threshold: >30 minutes to resolve simple updates
• Solution: Delete and recreate environment
• Prevention: Batch install packages, use environment.yml files

Resource Requirements

Time Investment

• Initial setup optimization: 15-30 minutes
• Learning mamba: 5 minutes (same commands)
• Environment recreation: 10-60 minutes depending on complexity

Storage Requirements

• Conda installation: 500MB base
• Micromamba: 5MB base
• Cache maintenance: Clean monthly to prevent multi-GB accumulation

Expertise Level

• Basic optimization: Beginner (config changes)
• Tool migration: Intermediate (mamba/micromamba)
• Advanced troubleshooting: Expert (environment debugging)