This series of patches help to make IO processing more common for path-latency prioritizer and make it more user-friendly. First, the SCSI-to-NVMe translations which was blamed broken has been removed since linux kernel 4.13, so that SG_IO IOCTL used in the reading is not supported. Instead, this patch drops sg_read method and uses direct IO reading both for NVMe device and SCSI device. Second, the original prio_args for prioritizer is like this: 20|10 which is somewhat unconvenient for user. Third patch drops it and use a syntax that similar to other prioritizers, for example : "base_num=5 io_num=10". PATCH 2/3 can be an independent patch. Those above are the two things being addressed by this series of patches. Note: Martin has pointed that standard deviation in the path-latency prioritizer is wrong. We are solving this but it will take more days to prove the correctness and verify this by testing. Thanks Guan Junxiong Junxiong Guan (3): multipath-tools: use direct IO for path latency prioritizer multipath-tools: move get_next_string to util multipath-tools: use user-friendly prio_args for path-latency libmultipath/prioritizers/path_latency.c | 156 +++++++++++++++++++++++-------- libmultipath/prioritizers/weightedpath.c | 10 +- libmultipath/util.c | 9 ++ libmultipath/util.h | 1 + multipath/multipath.conf.5 | 2 +- 5 files changed, 127 insertions(+), 51 deletions(-) -- 2.11.1 -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel