NVMe Storage Administration

NVMe Storage Administration

NVMe (Non-Volatile Memory Express) is the storage protocol used by modern SSDs that connect via PCIe. NVMe drives are fundamentally different from SATA SSDs — they have much lower latency, higher IOPS, and different device naming conventions. Most cloud VMs and modern bare-metal servers use NVMe storage, so understanding how to administer NVMe devices on Ubuntu is increasingly essential.

NVMe vs SATA SSD

NVMe device naming

NVMe device naming convention:
/dev/nvme0n1      ← Controller 0, Namespace 1 (the storage device)
/dev/nvme0n1p1    ← Controller 0, Namespace 1, Partition 1
/dev/nvme1n1      ← Controller 1, Namespace 1 (second NVMe device)

nvme0 = first NVMe controller (physical device)
n1    = namespace 1 (logical block address space — usually 1 per device)
p1    = partition 1 within the namespace

# List all NVMe devices
ls /dev/nvme*
lsblk | grep nvme

# NVMe tools
sudo apt install -y nvme-cli

# List all NVMe devices with details
sudo nvme list

nvme list output

Node             SN                   Model               Namespace Usage        Format
/dev/nvme0n1     S6XXNX0T800000      Samsung SSD 980 PRO 1         512.11  GB   512   B + 0 B

NVMe health monitoring

# Show NVMe SMART/health data
sudo nvme smart-log /dev/nvme0n1

nvme smart-log output (key fields)

Smart Log for NVME device:/dev/nvme0n1 namespace-id:ffffffff
critical_warning                    : 0     ← 0 = no warnings
temperature                         : 38 C  ← Operating temperature
available_spare                     : 100%  ← % spare blocks left
available_spare_threshold           : 10%   ← Alert below this
percentage_used                     : 0%    ← Drive wear (100% = end of life estimate)
data_units_read                     : 1,234,567
data_units_written                  : 2,345,678
power_on_hours                      : 1,234
unsafe_shutdowns                    : 5

# Get specific health field
sudo nvme smart-log /dev/nvme0n1 | grep "percentage_used"

# Check error log
sudo nvme error-log /dev/nvme0n1

# Verify data integrity (end-to-end data protection)
sudo nvme sanitize-log /dev/nvme0n1

NVMe performance tuning

# Check current I/O scheduler (NVMe should use 'none')
cat /sys/block/nvme0n1/queue/scheduler

Expected output for NVMe

[none] mq-deadline kyber bfq
# 'none' is already selected (in brackets) — correct for NVMe

# NVMe has its own internal queuing (thousands of queues vs 1 for SATA)
# The 'none' scheduler lets NVMe manage its own queue — correct setting

# Check queue depth
cat /sys/block/nvme0n1/queue/nr_requests    # Typically 1023 for NVMe

# Enable TRIM support for NVMe
# In /etc/fstab, add 'discard' option:
UUID=...  /data  ext4  defaults,noatime,discard  0  2

# Or schedule periodic TRIM (less write amplification)
sudo systemctl enable --now fstrim.timer
sudo systemctl status fstrim.timer

NVMe namespace management

# List namespaces on an NVMe controller
sudo nvme list-ns /dev/nvme0

# Show namespace details
sudo nvme id-ns /dev/nvme0n1

# Partitioning NVMe is the same as SATA — use fdisk or parted
sudo fdisk /dev/nvme0n1    # Same commands as with /dev/sdb
sudo parted -s /dev/nvme0n1 mklabel gpt mkpart primary ext4 0% 100%

# Format partition (note: nvme0n1p1, not nvme0n11)
sudo mkfs.ext4 /dev/nvme0n1p1

Conclusion

NVMe administration is largely the same as SATA disk administration — fdisk, parted, mkfs, mount, fstab — but with different device names (/dev/nvme0n1 instead of /dev/sda). Install nvme-cli for health monitoring via nvme smart-log. The I/O scheduler should be none for NVMe (the default). Enable TRIM with the discard mount option or use fstrim.timer for periodic TRIM. Monitor percentage_used in smart-log to track drive wear — replace the drive before it reaches 100%.