When copying files between two physical drives on Linux, the underlying mechanics differ significantly between commands:

# CP (Copy) example
cp -rpv /mnt/source/directory /mnt/destination/

# MV (Move) example when crossing filesystems
mv /mnt/source/file /mnt/destination/file

# RSYNC example
rsync -ah --progress /mnt/source/ /mnt/destination/

CP: Performs raw file copying with minimal overhead. Modern GNU cp implementations use efficient copy-on-write techniques.

MV: When operating within the same filesystem, mv is instantaneous (metadata-only operation). Across drives, it degenerates into cp+rm behavior.

RSYNC: Adds checksum verification and delta-transfer capabilities, creating additional overhead but providing resilience.

Testing with a 10GB directory containing mixed file sizes (1KB-2GB) on ext4 filesystems:

*When crossing filesystem boundaries

For maximum performance:

# Use parallel copying with GNU parallel
find /mnt/source -type f -print0 | parallel -0 -j8 cp {} /mnt/destination/{}

# Direct block device copying (when filesystems are identical)
dd if=/dev/sdb of=/dev/sdc bs=64K status=progress

• Use cp for simple, one-time transfers
• Choose rsync when you need verification or network capabilities
• Consider parallel or dd for raw speed in special cases
• Remember that mv offers no advantage when crossing filesystems

When transferring files between mounted drives on Linux, the underlying system calls and I/O patterns differ significantly between commands. Let's examine the technical behaviors:

# Sample disk-to-disk transfer commands
cp -rp /mnt/drive1/data /mnt/drive2/backup
rsync -ah --progress /mnt/drive1/data/ /mnt/drive2/backup/
mv /mnt/drive1/large_file.iso /mnt/drive2/

cp: Performs full read-write operations, including metadata preservation. Modern GNU cp uses copy_file_range() syscall for efficient copying.

mv: When source and destination are on different filesystems, it actually performs cp+rm internally. Only same-filesystem moves benefit from metadata-only operations.

rsync: Adds overhead for delta-transfer calculations, but shines in incremental sync scenarios with --inplace flag for direct overwrites.

In controlled tests with ext4 filesystems and 50GB file sets:

For maximum throughput:

# Use parallel copy with GNU parallel
find /mnt/drive1/data -type f -print0 | parallel -0 -j 8 cp {} /mnt/drive2/backup/

# Direct I/O bypassing page cache
dd if=/mnt/drive1/largefile of=/mnt/drive2/largefile bs=1M iflag=direct oflag=direct

• cp: Simple one-time transfers, when you need POSIX compliance
• mv: Testing if files are on same physical device (quick check)
• rsync: Network transfers, resume capability, checksum verification

For specialized use cases:

# pv for progress monitoring and throughput measurement
pv /mnt/drive1/bigfile > /mnt/drive2/bigfile

# mc (Midnight Commander) for interactive transfers
mc /mnt/drive1 /mnt/drive2