Faster IO_uring, BFQ + BLK-MQ Improvements Among The I/O Fun For Linux 5.12
The block subsystem and related storage changes were merged today for the in-development Linux 5.12 kernel.
The IO_uring changes for Linux 5.12 continue to be quite prominent for this very compelling feature of the Linux kernel. In particular, continued work on making IO_uring even faster. With request recycling and task_work optimizations, IO_uring with Linux 5.12 is now in the range of 10% to 20% faster for workloads that are mostly inline. IO_uring is also now fully under memcg protection, SQPOLL fixes, LOOKUP_CACHED support, and other clean-ups and optimizations.
Meanwhile the block changes include BFQ improvements, BLK-MQ scheduler improvements, and a variety of other improvements along with some other minor features like zoned write granularity.
Meanwhile the block drivers have a lot of NVMe updates as usual, a RAID5 fix for the MD code, BCache fixes, and even a floppy driver fix.
I'll have up some fresh NVMe SSD benchmarks and Linux file-system benchmarks from the 5.12 kernel once the merge window has passed.
The IO_uring changes for Linux 5.12 continue to be quite prominent for this very compelling feature of the Linux kernel. In particular, continued work on making IO_uring even faster. With request recycling and task_work optimizations, IO_uring with Linux 5.12 is now in the range of 10% to 20% faster for workloads that are mostly inline. IO_uring is also now fully under memcg protection, SQPOLL fixes, LOOKUP_CACHED support, and other clean-ups and optimizations.
Meanwhile the block changes include BFQ improvements, BLK-MQ scheduler improvements, and a variety of other improvements along with some other minor features like zoned write granularity.
Meanwhile the block drivers have a lot of NVMe updates as usual, a RAID5 fix for the MD code, BCache fixes, and even a floppy driver fix.
I'll have up some fresh NVMe SSD benchmarks and Linux file-system benchmarks from the 5.12 kernel once the merge window has passed.
5 Comments