This problems repeats on different installations again and again (many people reports it).
What I wounder is - why, if it is SO SIMPLE to allow O2CB to have a few heartbeat channels (at least few IP addresses), and it is so safe to do it (such change is 100% safe, it is not thing like _dont self fence if FS is in consistant state and we have not outstanding IO, ecven if we lost all connections_) - why OCFS team did not made such improvement during few years of development? It is 100% obvious, and 100% proven by many different clusters, that clustered system MUST (not SHOULD) have few heartbeat channels (and dont rely on L2 leyer such as interface bonding). In most cases, such change is 100% safe and simple. I can understand Oracle, which have heartbeat channel merged with data exchanged channel - in this case, using few channels can cause data to come out of order (but - Oracle support multi interfaces for css syncronization, at least in theory). OC2b heartbeat have not this problem, at least for heartbeat. So, why this simple thing takes so may time to be even proposed? I can bet, that one of 2 things will happen in next, say, 2 years: - OCFSv2 wil have multy interface feature OR - OCFS2 will die as a product (to be correct, O2CB, becasue OCFSv2 can work without it). PS. TCP/[EMAIL PROTECTED] have 30 - 45 seconds convergence time, which means that in any moment, you can have 30 - 45 seconds service interruption. Most TCP/IP protocols, even iSCSI, are well adjusted for this. OCFSv2 have this problem from the very first days. For now, this problem became one of OCFSv2 stoppers (if not killers). There is a very simple fix (multyinterface). Conclusion? _______________________________________________ Ocfs2-users mailing list [email protected] http://oss.oracle.com/mailman/listinfo/ocfs2-users
