What are the best practices of implementing Lustre in a virtualized environment (Cloud) specifically when there are no shared-disks (multipath) nor a remote ability to shut down a node (e.g. IPMI.)
The major issue would be setting up the MGS without the shared-disk MGT. There are discussions of using DRBD and some think I should setup an iSCSI target, perhaps also for Fencing. Similarly, Pacemaker with corosync has been discussed, but the options are enormous. It is important that there be no single point of failure and any outage be repairable as automatically as possible. Thoughts? Paul Hergt [email protected]
_______________________________________________ lustre-discuss mailing list [email protected] http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
