In NFS version 3 (NFSv3), the default rsize (read size) and wsize (write size) values are typically set to 32KB (32768 bytes) when not explicitly specified in the mount options. These values represent the maximum amount of data that can be transferred in a single read or write operation between the NFS client and server.

To check the actual values being used in your current NFS mount, run:

cat /proc/mounts | grep nfs

Or for more detailed NFS-specific information:

nfsstat -m

For read-heavy workloads, you might want to test various rsize values. Here's how to mount with custom sizes:

mount -t nfs -o rsize=65536,wsize=65536 {server_ip}:/home/{server_user}/{server_path} /home/{client_user}/{client_path}

Or in /etc/fstab:

{server_ip}:/home/{server_user}/{server_path} /home/{client_user}/{client_path} nfs rsize=65536,wsize=65536 0 0

Typical values to benchmark for read-heavy operations:

• 8192 (8KB) - Minimum useful value
• 32768 (32KB) - Default
• 65536 (64KB) - Common optimization
• 131072 (128KB) - For high bandwidth networks
• 262144 (256KB) - Maximum for most implementations

Use tools like dd or iozone to measure performance. Example read test:

dd if=/home/{client_user}/{client_path}/testfile of=/dev/null bs=1M count=1000

For more comprehensive testing with iozone:

iozone -a -i 0 -i 1 -s 1G -r 64k -f /mnt/nfs/testfile

Remember that optimal values depend on:

• Network MTU (1500 bytes standard Ethernet)
• TCP window scaling
• Server and client hardware capabilities
• Filesystem block sizes on both ends

If you experience timeouts or errors with large values, try adjusting these additional parameters:

mount -t nfs -o rsize=131072,wsize=131072,timeo=600,retrans=2 {server_ip}:/path /mnt

In NFSv3 implementations, the default read (rsize) and write (wsize) transfer sizes typically vary between different operating systems and NFS client implementations:

# Common defaults across platforms:
- Linux kernels: usually 32KB (32768 bytes)
- FreeBSD: 8KB (8192 bytes)
- Solaris: 32KB default
- AIX: 40KB (40960 bytes)

To verify your current NFS mount parameters:

# Check mounted NFS parameters
cat /proc/mounts | grep nfs

# Alternative method using nfsstat
nfsstat -m

# Sample output (truncated):
# 172.16.1.100:/export/data on /mnt/data type nfs (rw,relatime,vers=3,rsize=32768,wsize=32768,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,mountaddr=172.16.1.100,mountvers=3,mountport=892,mountproto=tcp,local_lock=none,addr=172.16.1.100)

For read-heavy workloads, consider these benchmarking approaches:

# Test sequential read performance with various rsize values
for size in 4096 8192 16384 32768 65536 131072; do
    mount -o remount,rsize=$size,wsize=$size ${server_ip}:/path /mnt/nfs
    echo "Testing rsize/wsize=$size"
    dd if=/mnt/nfs/largefile of=/dev/null bs=1M count=1024
done

# Measure IOPS with fio
cat > fio_test.ini <<EOF
[global]
ioengine=libaio
direct=1
runtime=60
size=1G

[readtest]
rw=randread
bs=4k
directory=/mnt/nfs
numjobs=4
EOF

fio fio_test.ini

Key factors affecting optimal transfer sizes:

• Network MTU (1500 bytes standard, 9000 for jumbo frames)
• Server and client hardware capabilities
• NFS server configuration (e.g., nfsd thread count)
• Workload characteristics (sequential vs random access)

# /etc/fstab entry with optimized values for read-heavy workload
{server_ip}:/home/{server_user}/{server_path}  /home/{client_user}/{client_path}  nfs  rw,rsize=65536,wsize=32768,hard,intr,tcp,timeo=600  0  0

# Mount command alternatives
mount -t nfs -o rsize=131072,wsize=65536 ${server_ip}:/path /mnt/nfs

Essential tools for ongoing optimization:

# Monitor NFS statistics
nfsiostat -d 5  # Similar to iostat but for NFS
nfsstat -c      # Client-side statistics

# Kernel parameters worth tuning (sysctl)
vm.dirty_ratio = 40
vm.dirty_background_ratio = 10
sunrpc.tcp_slot_table_entries = 128

When experimenting with large transfer sizes:

# Check for "fragmenting" messages in dmesg
dmesg | grep -i nfs

# Verify network MTU consistency
ping -M do -s 8972 ${server_ip}  # For 9000 byte MTU testing

# Important: Always test with actual workload patterns rather than synthetic benchmarks alone