F2FS In Linux 4.19 Will Fix Big Performance Issue For Multi-Threaded Reads
The Linux 4.19 kernel updates for the Flash-Friendly File-System (F2FS) should bring much faster performance for multi-threaded sequential reads -- as much as multiple times faster.
Two years ago F2FS dropped its write-pages lock on the basis it could improve multi-threading performance... 4KB writes across 32 threads went up from 25 to 28MB/s on some tests done on the developer's hardware. While it was a minor win for multi-threaded writes, it turns out dropping the write-pages lock took a major toll on the multi-threaded read performance. Now with Linux 4.19, that write-pages lock is being restored.
Jaegeuk Kim reverted the removal of the write-pages lock for this next kernel cycle in F2FS. With multi-threaded sequential reads the read throughput goes up from 185 MB/s to now 758 MB/s. Details in this commit for the staged code ahead of the Linux 4.19 merge window.
The F2FS file-system for Linux 4.19 is also getting a fault_type mount option and a range of other fixes and code improvements for further polishing this flash-focused file-system.
Two years ago F2FS dropped its write-pages lock on the basis it could improve multi-threading performance... 4KB writes across 32 threads went up from 25 to 28MB/s on some tests done on the developer's hardware. While it was a minor win for multi-threaded writes, it turns out dropping the write-pages lock took a major toll on the multi-threaded read performance. Now with Linux 4.19, that write-pages lock is being restored.
Jaegeuk Kim reverted the removal of the write-pages lock for this next kernel cycle in F2FS. With multi-threaded sequential reads the read throughput goes up from 185 MB/s to now 758 MB/s. Details in this commit for the staged code ahead of the Linux 4.19 merge window.
The F2FS file-system for Linux 4.19 is also getting a fault_type mount option and a range of other fixes and code improvements for further polishing this flash-focused file-system.
20 Comments