On 01/28/2016 09:50 AM, Wenwei Tao wrote: > 2016-01-27 17:44 GMT+08:00 Matias Bjørling <m...@lightnvm.io>: >> On 01/26/2016 01:33 PM, Wenwei Tao wrote: >>> When create a target, we specify the begin lunid and >>> the end lunid, and get the corresponding continuous >>> luns from media manager, if one of the luns is not free, >>> we failed to create the target, even if the device's >>> total free luns are enough. >>> >>> So add non-continuous lun target creation support, >>> thus we can improve the backend device's space utilization. >>> Signed-off-by: Wenwei Tao <ww.tao0...@gmail.com> >>> --- >>> Changes since v1: >>> -use NVM_FIXED instead NVM_C_FIXED in gennvm_get_lun >>> -add target creation flags check >>> -rebase to v4.5-rc1 >>> >>> drivers/lightnvm/core.c | 36 ++++--- >>> drivers/lightnvm/gennvm.c | 42 ++++++++- >>> drivers/lightnvm/rrpc.c | 215 >>> +++++++++++++++++++++++++++--------------- >>> drivers/lightnvm/rrpc.h | 6 +- >>> include/linux/lightnvm.h | 24 ++++- >>> include/uapi/linux/lightnvm.h | 3 + >>> 6 files changed, 229 insertions(+), 97 deletions(-) >>> >> >> Hi Wenwei, >> >> I did some digging on the patch and changed the interface to a >> reserve/release interface. I also removed the logic to dynamically >> select another lun than the one requested. >> >> A couple of questions: >> >> 1. The rrpc_lun->rev_lock and rev_trans_map change; this might be for >> another patch, and it isn't directly related to continuous mapping? > > rrpc_lun->rev_lock and rev_trans_map change is related to > non-continuous mapping, it's not directly related to continuous > mapping. > Put this change in another patch along with non-continuous mapping > support and this patch would be only add reserve/release thing, is > that your suggestion?
Yes, that would be great. Then we keep it separate. I'll like to do some benchmarks with the patch on and off, to see the performance difference. > >> 2. Instead of dynamically assigning new luns when not available, what >> about taking a list of lun ids instead? >> > > Seems you prefer user make the choice ? Yes, I want it to be deterministic. For example, if we do it dynamically, the user might first allocate 2-4, and then allocate 1-3 , which will actually allocate 0,1,5. Then later, a user tries to allocate on 0, and instead gets returned 6. It quickly makes it difficult to use. > But the target creation can still fail if one of the list lun ids is > not available although there may be enough free luns. Agree, the user would have to look up the free luns and then resubmit the target allocation.
