https://bugs.kde.org/show_bug.cgi?id=255306
Méven <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- Resolution|--- |FIXED Status|REOPENED |RESOLVED CC| |[email protected] --- Comment #34 from Méven <[email protected]> --- I measured this on current KIO master (2026-10-26, to be Frameworks 6.32): one 1 GiB file copied to a CIFS mount, compared with cp. The share is Samba over loopback, mounted with vers=3.1.1, and netem adds 2 ms each way and limits the link to 1 Gbit. KIO: 9.65 s cp: 9.45 s cp --preserve=mode,ownership,timestamps,xattr: 9.55 s At 1 Gbit the data alone takes 8.6 s, so KIO copies close to the line rate, within 2% of cp. The copy code has changed a lot since the reports here. Since Frameworks 5.88, a copy goes through copy_file_range(). Since 6.32, a copy between two filesystems, for example to a CIFS mount, uses sendfile() instead of a buffer in user space. The size of each copy step adapts to how long a call takes, and a network mount gets a writeback window. I'm closing this as fixed. If you still see a large difference with cp on Frameworks 6.32 or later, please reopen and add your mount options (from /proc/mounts) and both timings. -- You are receiving this mail because: You are watching all bug changes.
