Alain, Moullé Alain napsal(a): > Hi > Thanks Steven and Jan. > > OK, so "passive" mode seems to be more safer for now ... > > So questions about passive mode, just to clarify completely the behavior > (I think it could be useful for all users) : > > 1/ in the case both networks are up and working, is by default the ring > 0 used at first, and ring 1 as backup ? or is it a little bit random ? >
Passive is actually not very good name for mode (hence we took from paper). Packets are send in round-robin fashion via both interfaces (so packet 1 - ring 0, packet 2 - ring 1, packet 3 - ring 0, ...) > 2/ following failure on first ring, if the protocol switches to the > backup ring, and then the first ring comes up and working again, > does the protocol switches back to first ring, or does it remain on > backup ring until Yes. When failure happens packets are send only via "live" ring. When "dead" ring becomes active again (we are sending small messages to find out when it happens), everything backs to normal fully automatically. - > a. either if backup ring fails at its turn, or > b. stack HA is re-started on all nodes > > 3/ if in case 2/ it switches back on first ring, let's say we re-start > HA stack on only one node of the HA cluster, does it also switch back to > first ring , or does it remain on the ring > currently used by the other(s) node(s) > > Thanks > Alain > > Regards, Honza > Le 27/11/2013 16:54, Jan Friesse a écrit : >> Alain, >> passive mode is much better tested. Another big plus of passive is, that >> if one network becomes faulty, passive makes progress (one packet is >> send thru active device, another via faulty - this is not delivered but >> resend via active device, ...). Active RRP waits until enough failures >> and then marks device/ring failed. In this meantime, there is no >> progress (packet must be delivered via both interfaces). >> >> Moullé Alain napsal(a): >>> Hi, >>> >>> the man page of corosync.conf gives : >>> >>> "Active replication offers slightly lower latency from transmit to >>> delivery in faulty network environments but with less performance. >>> Passive replication may nearly double the speed of the totem protocol if >>> the protocol doesn’t become cpu bound" >>> >>> OK but knowing that, could someone give the pro & cons for passive mode, >>> and the pro & cons for active mode, >>> and/or how must we choose the real better mode for a HA cluster ? >>> >> Passive. It's only one supported. >> >>> Thanks a lot >>> Alain >>> >> Regards, >> Honza >> > _______________________________________________ Openais mailing list Openais@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/openais
