On Dec 15, 2005 16:24 +0000, John Rowe wrote:> * I will always buy disks in pairs, putting them onto different machines > connected by gigabit where they will act as replicas of each other. > > * If one machine of a pair goes down my users won''t notice (OST > failover).Lustre does not currently support "lustre level" OST replication. The only way to get this currently is to have block device-level replication with multi-host connectivity (RAID[15] + mutli-host connection like FC or firewire, MAYBE (G)NBD but untested). Latest word is that we are hoping to have lustre-level OST replication in late 2006 but no firm plans yet. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.
I''m hoping to use Lustre as a resilient way of serving home directories to my users. We have a heterogeneous collection of 32 and 64 bit machines running a RHEL4 clone so it''s not a collection of slaves with a single disk image (although I have one of those too). My hope is that: * I will always buy disks in pairs, putting them onto different machines connected by gigabit where they will act as replicas of each other. * If one machine of a pair goes down my users won''t notice (OST failover). * I can add replicated pairs to a single global file system and migrate out older disks * Performance will be at least as good as NFS without needing dedicated machines (people want to run jobs). * I can export the Lustre file system via samba or NFS for the odd legacy machine. * It will be very reliable and there will never be a time when Lustre just stops working. * A reboot will fix everything (for when I''m on holiday). * There will NEVER be an announcement like "Lustre 2.0 has been released, please upgrade every machines simultaneously". Is this realistic or have I got the wrong animal? Thanks. John