>>> Andrei Borzenkov <[email protected]> schrieb am 10.08.2015 um 08:38 in Nachricht <caa91j0ua8ow+bcfu60jbb_lpegoy3ar-8j60y69znp6zuks...@mail.gmail.com>: > On Mon, Aug 10, 2015 at 9:30 AM, Ulrich Windl > <[email protected]> wrote: >>>>> Steven Dake <[email protected]> schrieb am 09.08.2015 um 01:02 in > Nachricht >> <CAPwfPsjd_SGrC923OUU=q4leuki_4nkexucrzat8y7cix8c...@mail.gmail.com>: >>> Be careful with bonding - there are several different operational modes and >>> only one works. We tested bonding extensively at RHT with corosync and >>> came to the conclusion that only one bonding mode was reliable with >>> corosync. I don't recall what the bonding mode # was specifically, perhaps >>> Honza could read the documentation and report. >> >> Hi! >> >> I hope there won't be another message saying only one type of NIC was found > to be working with Corosync ;-) > > My experience is that if higher level protocol natively supports > multiple links with proper failure detection and failover, using lower > level aggregation just makes things worse. I'd rather see proper > support for multiple links on corosync level. > >> Before making statements like the one above, precisely list the requirements > that corosync has, preferably referencing the bugs that lead to those > requirements. >> My basic idea is: If some UDP protocol like NFS works with a specific > network connection, why shouldn't corosync work with a connection of equal > quality? > > Because different protocols have different requirements? You can lose > lot of NFS datagrams it will try recover indefinitely if suitably > configured; but corosync likely have much stricter requirements and > may not tolerate connectivity loss during switch between links.
The point is that NFS can be fitted to your needs (timeouts, soft vs. hard, etc.) Corosync also has a bucnh of parameters. My idea is that one could fit corosync also to one's needs. Regards, Ulrich [...] _______________________________________________ discuss mailing list [email protected] http://lists.corosync.org/mailman/listinfo/discuss
