Messages are sent to all replicas involved in there request at the same time. All nodes in the cluster must be able to communicate to all other nodes.
The coordinator the client is talking to, the local coordinator, groups messages (for one read/mutation) to be sent to remove data centres and only sends a single message to one node in each remote DC. In each remote DC the receiving node then acts as a remote coordinator which forwards the messages to the other replicas. All replicas in any DC (local and remotes) reply directly back to the local DC. Cheers ----------------- Aaron Morton Freelance Cassandra Consultant New Zealand @aaronmorton http://www.thelastpickle.com On 29/04/2013, at 11:47 PM, Sergey Naumov <sknau...@gmail.com> wrote: > Hello. > > I would like to know whether updates are propagated from local DC to remote > DCs simultaneously (so All-to-All network connections are preferable) or > Cassandra can somehow determine nearest DCs and send updates only to them (so > these nearest DCs have to propagate updates further)? > Is there some optimizations for multiple DCs placed sequentially on a single > link like DC1 <-> DC2 <-> ... <-> DCn? > > Thanks in advance, > Sergey Naumov.
