It would be an interesting exercise though. Depending on network
layout (no cluster network) the client could multicast to all
replicas
and potentially reduce latency by half. I suspect that the client
participating in the replication goes against the internal workings
of
ceph though and would be a major rework...
Having the client participate directly goes much further than I was
thinking (I was targeting more the server-side LAN).
I'd like to know how small block writes are replicated - maybe I should
analyse the traffic and find out.
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
