NFS as cluster storage - the simple shared filesystem · Linux

NFS (Network File System) is the standard shared filesystem for Linux clusters. It is simple, well-supported, and works across distributions. This lesson covers how to use it.

What NFS does

NFS allows multiple hosts to mount the same filesystem over the network:

NFS server (10.0.0.50)
  └── /srv/shared  (exported)

NFS client 1 (10.0.0.10)
  └── mount 10.0.0.50:/srv/shared /mnt/shared

NFS client 2 (10.0.0.11)
  └── mount 10.0.0.50:/srv/shared /mnt/shared

Both clients see the same files. The server is the source of truth.

NFS versions

NFSv3: widely supported, no encryption, no Kerberos.
NFSv4: modern, supports Kerberos, encryption, locking.
NFSv4.1: pNFS (parallel NFS) for high throughput.
NFSv4.2: latest.

For new deployments, use NFSv4.2.

Configure NFS server

# Install
sudo apt install nfs-kernel-server

# Create the export directory
sudo mkdir -p /srv/shared
sudo chown nobody:nogroup /srv/shared

Then declare the export. This is a configuration file, not a command — pasting the line below into a shell is a syntax error, because ( and ) are shell metacharacters:

# /etc/exports
/srv/shared 10.0.0.0/24(rw,sync,no_subtree_check)

Apply and confirm it:

sudo exportfs -ra          # re-read /etc/exports
sudo exportfs -v           # what is actually exported, with resolved options

exportfs -ra reports syntax errors on stderr and keeps the previous table, so read its output rather than assuming a silent run means success.

/etc/exports options:

rw / ro: read-write or read-only.
sync / async: synchronous or asynchronous writes.
no_subtree_check: disable subtree checking (faster, but less safe).
no_root_squash: do not map root to nobody (security risk; use carefully).
sec=sys / sec=krb5: security flavor (sys for default, krb5 for Kerberos).

Apply:

sudo exportfs -a

Configure NFS client

# Install
sudo apt install nfs-common

# Mount
sudo mount 10.0.0.50:/srv/shared /mnt/shared

Or in /etc/fstab:

10.0.0.50:/srv/shared /mnt/shared nfs defaults,_netdev 0 0

_netdev waits for the network to be up before mounting.

NFS performance

NFS performance depends on:

Network latency.
Server disk speed.
NFS version (v4 is faster than v3).
Sync vs async writes.
Block size (rsize and wsize mount options).

For high throughput:

10.0.0.50:/srv/shared /mnt/shared nfs rsize=1048576,wsize=1048576,hard 0 0

Production considerations

Single point of failure: the NFS server is. Use a clustered filesystem (GlusterFS, CephFS) for HA.
Network dependency: NFS over a slow network is slow.
Locking: NFSv4 has better locking than NFSv3 but is not as good as local filesystems.
Security: NFSv3 has no authentication. Use NFSv4 with Kerberos for security.

For cluster applications, prefer stateless where possible. NFS is good for static assets and small files.