Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 04/24/2011 12:59 AM, Patrick Egloff wrote: > Hi and thanks for the answer ! > > I got several pm urging me NOT to use active/active and OCFS2. > > A more simpler active/passive and no OCFS2 would be the best choice.... > Too many things could go wrong with OCFS2 and active/active + MySQL. > > But you fully understood my configuration and thanks for your help. > My drbd.conf is almost like the one you sent me. > > But in my case, i must have another problem.... it's not working. > > One more question. I have 2 ethernet ports. eth1 is used to link both > boxes together. > Should i use for DRBD + Heartbeat a different IP address and class than > on eth0 which is on the LAN ? > > Patrick Hi Patrick, OCFS2 and GFS2 require cluster locking, which comes with a fair amount of overhead. Primary/Secondary DRBD with a "normal" filesystem like ext3 will certainly be faster, but in Secondary, you can not access the Secondary resource at all. Along with cluster locking comes the requirement for a cluster lock manager, and in turn, a cluster proper. I am not sure if heartbeat provides this, but I would have to assume so. Seeing as you have a cluster, than the additional overhead of cluster locking should, itself, be trivial. If your concern is performance though, rather than overhead, stick with Primary/Secondary. "It's not working" -> Needs more details. :) There is no requirement for DRBD, cluster traffic and LAN activity to be segregated. However, for security reasons, you would want to enable encryption if you did share the LAN port. That said, yes, segregate cluster, DRBD and LAN networks. If LAN (internet-facing) is separate from DRBD and cluster, you can safely avoid encryption and gain some healthy performance benefits. Further, I like to dedicate a link to DRBD so that periods of high disk activity don't put at risk cluster communications. Again, not knowing heartbeat, I have to speak from a RHCS perspective. There, if anything interrupts or delays packet delivery for more than 300ms*10, the cluster could partition and fences will start flying. Given the relative trivial expense of network cards, I always recommend three separate networks; Internet Facing, Storage and Back-Channel (cluster comms + live migrations when clustering VMs). -- Digimer E-Mail: digimer at alteeve.com AN!Whitepapers: http://alteeve.com Node Assassin: http://nodeassassin.org