Hi Guoqing Yes, near is the default, but some have suggested that far be made the default. Far has about double the reading speed as near, while writing speed is only a little less. https://marc.info/?l=linux-raid&m=141104637932757 So a number of people have said that "far" has the best overall performance, but then I think that "far" is not so well known in the community, still 10 years after its implementation in the kernel. I was the inventor of the "far" layout. here are my ideas for writing grow support for raid10 far layout (at the end) https://marc.info/?l=linux-raid&m=130288632726912&w=2 Best regards keld On Wed, Oct 10, 2018 at 09:37:49AM +0800, Guoqing Jiang wrote: > Hi Keld, > > On 10/9/18 8:25 PM, keld@xxxxxxxxxx wrote: > >Hi > > > >Are you thinking of also doing enhanced support for plain raid10? > >For example raid10 far layout lacks growing support. > > Yes, I am aware of that, this patchset is for clustered raid10 which only > support near layout so far. > > >I once wrote up some ideas on how to implement growing support for raid10 > >far layout - > >it was posted on this list, but if you are interested I can send it again. > > Thanks, please send it to me or let me know the key word then I can search > it. > > BTW: I am curious if there are lots of people want to grow raid10 with far > layout, after all, the layout is n2 by default. > > Regards, > Guoqing > > > > >best regards > >keld > > > >On Tue, Oct 09, 2018 at 10:59:41AM +0800, Guoqing Jiang wrote: > >>Hi, > >> > >>This series aim to change the capacity for clustered raid10 > >>just like normal raid10. > >> > >>The first patch ensure all nodes have the same bitmap size > >>before reshape. After that, master node broadcasts RESYNCING > >>msg in reshape_request and slave nodes update reshape_position, > >>this is achieved by the second patch. > >> > >>The 3rd patch addes sanity check for resync_info_get, and > >>other patches are introduced to handle different kinds of > >>cases which happened during grow clustered raid10. And the > >>last patch cleans up suspend_info which is not necessary. > >> > >>Thanks, > >>Guoqing > >> > >>Guoqing Jiang (8): > >> md-cluster/raid10: resize all the bitmaps before start reshape > >> md-cluster/raid10: support add disk under grow mode > >> md-cluster: introduce resync_info_get interface for sanity check > >> md-cluster/raid10: call update_size in md_reap_sync_thread > >> md-cluster/raid10: don't call remove_and_add_spares during reshaping > >> stage > >> md-cluster/bitmap: don't call bitmap_sync_with_cluster during > >> reshaping stage > >> md-cluster: send BITMAP_NEEDS_SYNC message if reshaping is interrupted > >> md-cluster: remove suspend_info > >> > >> drivers/md/md-cluster.c | 234 > >> ++++++++++++++++++++++++++++++++---------------- > >> drivers/md/md-cluster.h | 2 + > >> drivers/md/md.c | 65 ++++++++++++-- > >> drivers/md/md.h | 1 + > >> drivers/md/raid10.c | 81 ++++++++++++++++- > >> 5 files changed, 297 insertions(+), 86 deletions(-) > >> > >>-- > >>2.12.3
