MD In Linux 4.10 Adds RAID5 Writeback Cache Feature, FAILFAST

The Linux MD/RAID code was updated today in Git for the Linux 4.10 kernel with new functionality.

One of the new MD additions for Linux 4.10 is a RAID5 write-back cache feature. This aggregates writes to make a full stripe write and reduce read-modify-writes. MD maintainer Shaohua Li reported this feature should be good for workloads doing sequential writes followed by fsync. For now though this RAID5 cache feature is considered experimental and disabled by default. The r5c_journal_mode sysfs entry is used for setting the write-back or write-through cache mode.

"Failfast" support has been added. The failfast flag fails I/O to broken RAID disks to improve the latency. More details on the FAILFAST work via this earlier patch series.

Also new to the MD code for Linux 4.10 is AVX2 instructions are supported now along the RAID6 parity calculation path. Advanced Vector Extensions 2 are supported with Intel Haswell and newer or AMD Carrizo/Zen.

Various other MD changes -- including bug-fixes -- can also be found in this pull. Check it out for more details.