StorageCraft

Learn and operate the data path—from first principles to production-minded practice.

Interactive labs, transparent engineering models, and visual deep dives spanning storage internals, distributed data, and LLM infrastructure.

Why StorageCraft exists

The storage layer is usually invisible—until latency spikes, a disk fails, compaction stalls, or an LLM server runs out of memory. StorageCraft exposes those mechanisms through experiments you can operate and models you can inspect.

New here? Start the ordered StorageCraft Foundations: Zero-to-SAN Engineer program. Use Learn, Practice, Challenge, and Assess modes to progress from storage vocabulary to fabric operations and incident response.

Explore	Operate	Measure
Canonical systems explanations	Browser-based failure labs	Capacity and memory calculators
Storage × AI learning paths	Colored terminal workflows	Explicit formulas and assumptions
Production trade-off checklists	Shareable, zero-install demos	Reproducible benchmarks—next

Try it in 60 seconds

Browser lab

Open the Universal RAID Planner to compare RAID 0–6, nested RAID, and RAID-Z. Then use the focused RAID-5 Visualizer to write stripes, fail a disk, and watch XOR reconstruction.

Colored CLI

git clone https://github.com/manmeetnain/storagecraft.git
cd storagecraft
npm install
npm run craft -- raid --level 60 --disks 16 --size 8 --groups 2
npm run craft -- raid --compare --disks 12 --size 8 --groups 2
npm run craft -- kv --layers 32 --heads 8 --dim 128 --tokens 8192 --bytes 2

The CLI uses no additional runtime dependencies and respects the NO_COLOR convention.

Python toolkit

PYTHONPATH=python python3 -m storagecraft_tools.cli raid --level 6 --disks 8 --size 4
PYTHONPATH=python python3 -m storagecraft_tools.cli raid --compare --disks 12 --size 8 --groups 2
PYTHONPATH=python python3 -m storagecraft_tools.cli learn

The dependency-free Python package exposes typed, tested functions for notebooks and automation. It covers JBOD; RAID 0–6, 01, 10, 50, and 60; RAID-Z1/Z2/Z3; erasure coding; NVMe queues; RAG storage; and GPU memory. Start with the executable quick-start notebook, and use the browser planner for visual comparison.

Guided learning program

Stage	Capability	Practice
1–3	Storage types, performance, RAID, erasure coding, backup, RPO/RTO	RAID and erasure-coding labs
4–6	SAN paths, FC identity/login, zoning and masking	Network Academy Learn + Practice modes
7–8	iSCSI, multipathing, daily operations, incident handling	Failure labs, runbooks, Challenge mode
9	Explain, operate, troubleshoot, and communicate safely	Practical assessment

The program is backed by learning/catalog.json, validated in CI, and supported by a progress and evidence dashboard, glossary, operations runbooks, and instructor kit.

Current learning capsules

Path	Core question	Artifact
Write Amplification	Why does one logical write become many physical writes?	Deep dive
Write-Ahead Log	How is a mutation made crash-safe?	Deep dive
Copy-on-Write	How do immutable updates enable snapshots?	Deep dive
Erasure Coding	How is durability traded against capacity and repair cost?	Deep dive
KV Cache	Why does inference memory grow with live tokens?	AI-infra deep dive
GPU Memory Anatomy	What competes for accelerator memory?	Sizing workflow
RAID-5	How does distributed parity recover missing data?	Interactive lab
Universal RAID Planner	How do standard, nested, and RAID-Z layouts compare?	Interactive planner + CLI
Erasure Coding Lab	How does k+m coding trade capacity for recoverability?	Interactive failure lab + CLI
Write Amplification Explorer	How does one logical write multiply through the full stack?	Interactive pipeline + CLI
LSM-Tree Compaction Lab	How do leveled and tiered policies trade writes, reads, and space?	Interactive policy lab + CLI
GPU Memory Planner	Will an LLM serving configuration fit per GPU?	Interactive inference planner + CLI
RAG Storage Sizer	What is the full source, chunk, vector, index, and replica footprint?	Interactive AI-data planner + CLI
AI Data Path Lab	Which storage, fabric, CPU, or accelerator stage limits an AI workload?	Interactive bottleneck lab + CLI
NVMe Queue Lab	When do queue count and depth expose performance versus add latency?	Interactive queue lab + CLI
SAN Failure-Domain Lab	Will storage remain reachable—and recover before the workload times out?	Interactive resilience lab + CLI
Storage Network Academy	How can a new hire safely practice FC switching and iSCSI operations?	Stateful terminal academy + nine inspect-to-rollback workflows
Manmeet AI Command Center	Which AI platform, capability, skill, connector, shortcut, or prompt pattern fits the job?	Multi-platform deep-dive guide and interactive prompt console

Roadmap

• RAID failure and rebuild lab
• Universal RAID planner and RAID/KV-cache CLI models
• Storage internals and AI-infrastructure foundation
• Erasure Coding Lab v1
• Write Amplification Explorer v1
• GPU Memory Planner
• RAG Storage Sizer
• AI Data Path Lab
• Storage Network Academy: Brocade-style FC, Cisco MDS-style FC, and iSCSI
• Storage Network Academy v2: dual fabrics, multi-switch operations, incidents, configuration safety, governance, certification, and desktop/mobile E2E (execution contract)
• Academy operational workflows: inspect, diagnose, configure, verify, and rollback across Brocade, Cisco MDS, and iSCSI
• Zero-to-SAN curriculum, glossary, operational runbooks, and instructor kit
• Learner progress, milestone badges, and portable evidence export
• Full Python model parity foundation and executable notebook
• Reproducible benchmark capsule format

Vote for the next capsule through a topic proposal, ask a learning question in Discussions, or follow versioned changes in the changelog.

Quality contract

Every flagship capsule should include a learning objective, explicit assumptions, a visual or executable model, failure behavior, production limitations, automated validation, and a contribution entry point.

npm run check
npm run craft -- doctor

Contribute

Storage engineers, ML-systems engineers, database practitioners, educators, technical writers, and accessible-design contributors are welcome. Start with CONTRIBUTING.md, open a focused proposal, or improve an explanation with a primary source.

Maintainer

Built by Manmeet Nain—Enterprise Storage Engineer working across IBM, EMC, Brocade, Cisco, SAN/NAS, and AI infrastructure in India.

If StorageCraft helps your work, star the repository, share a lab, or sponsor its development.

---

_{Educational engineering software · Validate assumptions before production use · MIT}