I'm wondering why the dm block devices have a request queue at all, and more importantly why they use the default cfq scheduler. It seems to me that the physical disk should take care of the scheduling, and not have that work duplicated at each layer of dm stacked on top. It wonder though, where should the scheduler policy be set? In libdevmapper, or by a system udev rule? -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel