On Jun 20, 2014, at 9:47 AM, David Goldsmith <dgoldsmith at sans.org> wrote: > Here is a sample of running iostat on a server that has a LUN from a SAN with multiple paths. I am specifying a device list that just grabs the bits related to the multi path device: > > $ iostat -dxkt 1 2 sdf sdg sdh sdi dm-7 dm-8 dm-9 […] > For the rkB/s and wkB/s columns, the numbers for sdf and sdg add up to equal the numbers for dm-9. > > But for the first four columns: > > Column sdf/sdg dm-9 > ====== ============= ==== > rrqm/s 0.0 + 0.0 = 0.0 0 > wrqm/s 54.0 +13.0 = 67.0 0 Very different > r/s 7.0 + 1.0 = 8.0 8.0 > w/s 48.0 + 26.0 = 74.0 141.0 Very different > > So read data matches and write data diverges > > > Which numbers should I go with? The physical devices or the logical device? I may be wrong but I believe the difference you are seeing is due to write coalescing happening between the DM layer and the underlying SCSI layer reducing the number of write requests even though the same amount of data is read/written — Mark Tinberg mtinberg at wisc.edu