i now understand what you mean. i saw that async write operation and i
do not realize that is what u think write only.

so base on the patch, since i already know exactly that one device will
never be read. a simpler way might be

1) modify mdadm to allow one device to be configured as write only
2) add a write only flag in device info and check the flag when do read
balance.

so no need to update the time at all if no persistent bitmap purpose.

i saw the 2.4.x kernel raid has write only flag while dropped in 2.6.x.
i guess that is for reconstruction. but we can make use of that for this
purpose as well.

"write only" can be part of the async write or async replication. but i
think raid1 code should be able to allow write only to be a standalone
feature so others may have their own async write solution.

ming



On Wed, 2004-11-10 at 10:56, Paul Clements wrote:

Ming Zhang wrote:

quoted

On Wed, 2004-11-10 at 10:05, Paul Clements wrote:

quoted

Ming Zhang wrote:

quoted

is there any technical issue that lead not to be implemented?

No. In fact, it has been implemented. Both Peter T. Breuer and I have 
come up with slightly different patches to implement the write only 
functionality.

can u send me the patch then i can have a look?

my patches are here:

http://parisc-linux.org/~jejb/md_bitmap/

you're looking for the async write portion of the code, which has the 
write-only functionality in it (later patches do not contain that code, 
so look at the older ones)

check the mailing list archives for discussion

Peter's patch is here:

http://marc.theaimsgroup.com/?l=linux-raid&m=109238533827986&w=2


--
Paul