On Wed, 5 Nov 2014, jianpeng wrote:
> Dong Yuan <yuandong <at> unitedstack.com> writes:
>
> >
> > Hi Sage,
> >
> > I am now working with the BP osd: update Transaction encoding, but the
> > Transaction::get_data_alignment make me confused.
> >
> > This method give the alignment which is used by FileJournal to do
> > better buffer build. It calculate the alignment by largest_data_off
> > and get_data_offset() while the first is an offset of some object and
> > the second is an offset of the transaction encode result. I am not
> > sure there is any reason to do calculation between them.
> >
> > The code works fine, probably because any result is fine for
> > Transaction::get_data_alignment, while
> > FileJournal::prepare_single_write can use any alignment value to build
> > logical bufferlist.
> >
> > Can you give me some explanation? Thank you.
> >
>
> Hi, i send a pull request to reduce the memcopy cause by unalignment.
> https://github.com/ceph/ceph/pull/2803
> I think get_data_offset() should not care largest_data_off.
I finally got to testing this and it looks good. Pulled the patches
manually into master. Sorry it took so long!
For the last two MOSDSubOp, I think those changes should be done as part
of this ticket
http://tracker.ceph.com/issues/9961
which will replace the common client replication op to use a fresh message
type that is properly optimized. Jianpeng, is this something you might be
interested in working on? We talked with Haomai about it during CDS but I
think he and Yuan Dong will be occupied with the ObjectStore::Transaction
encoding first, and I would really like to see this done in time for
hammer (jan 1f freeze).
Thanks!
sage
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
