[ClusterLabs] corosync in multicast mode produces lots of unicast traffic
Hi! I have a Proxmox cluster which uses Corosync as cluster engine. Corosync uses the "default" multicast configuration. Nevertheless, using tcpdump I see much more packets sent by corosync using unicast between the node members than multicast packets. Is this normal behavior? If yes, please point me to some documentation. If no, does this mean I have some broken configuration? Everthing seems to work fine with my setup. Thanks Klaus ___ Users mailing list: Users@clusterlabs.org https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
Re: [ClusterLabs] corosync in multicast mode produces lots of unicast traffic
Klaus, Hi! I have a Proxmox cluster which uses Corosync as cluster engine. Corosync uses the "default" multicast configuration. Nevertheless, using tcpdump I see much more packets sent by corosync using unicast between the node members than multicast packets. Is this normal behavior? If yes, please point me to some documentation. It really depends. If cluster is quiet (no configuration changes) so it basically only heartbeats then it's pretty normal that unicast trafic (used for heartbeats) is bigger than multicast one. If no, does this mean I have some broken configuration? Everthing seems to work fine with my setup. So then you don't have to worry :) Regards, Honza Thanks Klaus ___ Users mailing list: Users@clusterlabs.org https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org ___ Users mailing list: Users@clusterlabs.org https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
Re: [ClusterLabs] corosync in multicast mode produces lots of unicast traffic
Hi Jan! Thanks for your answer. >> I have a Proxmox cluster which uses Corosync as cluster engine. >> Corosync uses the "default" multicast configuration. Nevertheless, >> using tcpdump I see much more packets sent by corosync using unicast >> between the node members than multicast packets. >> >> Is this normal behavior? If yes, please point me to some documentation. > > It really depends. If cluster is quiet (no configuration changes) so it > basically only heartbeats then it's pretty normal that unicast trafic > (used for heartbeats) is bigger than multicast one. Now I am confused. I thought that corosync uses totem and the totem protocol has implicit keep-alive by passing the token between the ring members. So, even if there are no messages, the token is passed on in the ring giving implicit keep-alive. And all this is done using multicast. So, is my understanding wrong and corosync uses totem only for message delivery and there is an additonal "heartbeat" feature which sends unicast keep-alive to all the known members? Thanks Klaus ___ Users mailing list: Users@clusterlabs.org https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
Re: [ClusterLabs] corosync in multicast mode produces lots of unicast traffic
Klaus, Hi Jan! Thanks for your answer. I have a Proxmox cluster which uses Corosync as cluster engine. Corosync uses the "default" multicast configuration. Nevertheless, using tcpdump I see much more packets sent by corosync using unicast between the node members than multicast packets. Is this normal behavior? If yes, please point me to some documentation. It really depends. If cluster is quiet (no configuration changes) so it basically only heartbeats then it's pretty normal that unicast trafic (used for heartbeats) is bigger than multicast one. Now I am confused. I thought that corosync uses totem and the totem protocol has implicit keep-alive by passing the token between the ring members. So, even if there are no messages, the token is passed on in the ring giving implicit keep-alive. And all this is done using multicast. Yes, it's exactly as you've wrote but these messages uses unicast. What I called heartbeat is exactly this token passing. Multicast is used only for sending regular messages. So if the cluster is quiet, only token passed between nodes using unicast. Regards, Honza So, is my understanding wrong and corosync uses totem only for message delivery and there is an additonal "heartbeat" feature which sends unicast keep-alive to all the known members? Thanks Klaus ___ Users mailing list: Users@clusterlabs.org https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org