[linux-lvm] dmraid - where is the raid done?
Bryn M. Reeves
bmr at redhat.com
Mon Nov 3 11:05:36 UTC 2008
>> Wrong list - that's for the Linux MD RAID subsystem (mdadm, /dev/md* and
>> friends). You're looking for ataraid-list:
>> Which is the place for discussion of dmraid development.
> Evening Byrn,
> I am aware that the dm-raid list was the best one if he had a dm-raid
> specific question. However from what I read it sounded like he was
> interested in knowing what the differences were between ata/dm-raid
> and md (software) raid. Being a lurker on both lists I felt that the
> folks over at md would be a better choice.
He was asking where the RAID is done in dmraid. I think the response of
the MD folks to that question based on past experience, would be either
very general or something like: "Huh? Go ask the device-mapper guys" ;)
If you're using dmraid to activate volumes on an ata RAID / fakeraid
adapter then all the RAID work is being done in the kernel by device-mapper.
When dmraid activates a volume, it reads the metadata from the disk that
describes the array and translates that into a device-mapper table,
usually using some combination of the linear, dm-raid1, and dm-raid45
targets. These are the kernel modules that implement the RAID algorithms
(dm-raid1 uses the kernel thread kcopyd to do the actual work - you'll
see this process consuming some resources on a system with a busy mirror).
For more information on the targets and the tables that are used to
configure them see Alasdairs talks on LVM/device-mapper here:
More information about the linux-lvm