5.3a LVM concepts: Physical Volumes (PV), Volume Groups (VG), Logical Volumes (LV)¶

🌐 Context Introduction¶

When managing AI infrastructure, storage flexibility is critical. AI workloads often require dynamic storage resizing, snapshots, and efficient disk space utilization. Logical Volume Manager (LVM) is a storage management layer that sits between the physical disks and the file system, allowing engineers to create flexible, resizable storage pools.

Think of LVM as a way to combine multiple physical hard drives into one large virtual pool, then carve out custom-sized chunks from that pool as needed — without worrying about the underlying physical disk boundaries.

⚙️ What is LVM?¶

LVM abstracts physical storage into logical units. Instead of working directly with hard disk partitions (like /dev/sda1), LVM introduces three key building blocks that work together:

• Physical Volumes (PV) — the raw disks or partitions
• Volume Groups (VG) — the pool of combined storage
• Logical Volumes (LV) — the virtual partitions you actually use

📦 Physical Volumes (PV) — The Raw Building Blocks¶

A Physical Volume is any block device that LVM can use — typically a hard disk partition (e.g., /dev/sdb1) or an entire disk (e.g., /dev/sdc).

Key characteristics: - A PV is created by initializing a disk or partition with LVM metadata - Each PV is divided into small chunks called Physical Extents (PE) — typically 4 MB in size - Multiple PVs can be added to a single Volume Group

Think of PVs as individual bricks you will use to build a wall.

🗄️ Volume Groups (VG) — The Storage Pool¶

A Volume Group is a collection of one or more Physical Volumes. It acts as a unified storage pool from which Logical Volumes can be allocated.

Key characteristics: - A VG can span multiple disks of different sizes - The total capacity of a VG equals the sum of all its member PVs - You can add or remove PVs from a VG without disrupting existing data (with some limitations) - The VG manages the allocation of Physical Extents to Logical Volumes

Think of a VG as a large water tank — you can add more water (disks) to it, and then draw water (storage) from it as needed.

💾 Logical Volumes (LV) — The Usable Storage¶

A Logical Volume is a virtual block device created from the free space within a Volume Group. This is what you format with a file system (like ext4 or xfs) and mount for use.

Key characteristics: - LVs can be resized (grown or shrunk) dynamically - You can create snapshots of LVs for backups or testing - LVs are accessed via device paths like /dev/vg_name/lv_name - Multiple LVs can exist within a single VG

Think of LVs as buckets you fill from the water tank — you can make them bigger, smaller, or take a snapshot of what's inside.

📊 Visual Representation: Logical Volume Manager (LVM) Architecture¶

This diagram maps the structural layers of LVM: physical disks are grouped into a virtual volume group, which is then carved into logical volumes.

🛠️ How They Work Together — The LVM Stack¶

The relationship between PV, VG, and LV is a three-layer hierarchy:

1. Bottom layer: Physical disks or partitions become Physical Volumes (PV)
2. Middle layer: PVs are combined into a Volume Group (VG)
3. Top layer: From the VG, you create Logical Volumes (LV)

Visual flow: - Disk → PV → VG → LV → File System → Mount Point

📊 Comparison Table: PV vs VG vs LV¶

🕵️ Why LVM Matters for AI Infrastructure¶

AI workloads have unique storage demands that LVM handles well:

• Dynamic growth: Training datasets can grow unexpectedly. With LVM, you can expand a Logical Volume without unmounting or rebooting.
• Snapshots: Before running a risky experiment or model update, you can take a read-only snapshot of the LV for quick rollback.
• Disk replacement: If a physical disk fails, you can add a new PV to the VG and migrate data without downtime.
• Pooling: Combine multiple smaller disks into one large volume group for massive AI datasets.

✅ Key Takeaways for New Engineers¶

• PV = raw disk ready for LVM
• VG = pool of storage from multiple PVs
• LV = flexible virtual disk you actually use
• LVM separates physical hardware from logical storage — giving you freedom to resize, snapshot, and reorganize storage without touching hardware
• For AI infrastructure, LVM is a foundational skill for managing growing datasets and model storage efficiently

📘 Common LVM Workflow (Conceptual)¶

1. Identify available disks (e.g., /dev/sdb, /dev/sdc)
2. Initialize them as Physical Volumes
3. Create a Volume Group from those PVs
4. Create a Logical Volume from the VG's free space
5. Format the LV with a file system
6. Mount the LV to a directory (e.g., /data)
7. Use the mount point for AI datasets or model storage

💡 Pro Tip: Always check available space in your Volume Group before creating a new Logical Volume. Running out of space in a VG is like trying to fill a bucket from an empty tank.