Brinon Philippe wrote: > After installation and test of the version 1.2.6, I noted a strong decrease > of the CPU activity on the master node and a lighter decrease of the CPU > activity on each of the 10 slave nodes. > > Consequently, there is no need now to try to limit the communication activity. > FYI, I had a chat with Jan about this issue when he was visiting Toronto last week.
Two interesting factors emerged that I hadn't previously realized: 1. If there is a sequence of SYNC events to be processed, the confirmations go back at the end, so that if there was, for instance, a backlog of 50 events that get processed at once, only one confirmation need return for the final SYNC event. Thus, this is less expensive than I thought. 2. We think it may be possible to eliminate altogether the generation of SYNC events on all nodes that are not the origin for a set. This would reduce further the (not overly heavy) burden of generating and processing SYNCs coming from other nodes. That may improve things further (albeit only marginally). _______________________________________________ Slony1-general mailing list [email protected] http://gborg.postgresql.org/mailman/listinfo/slony1-general
