On 2010-05-13 15:43, Burnash, James wrote: Hi James, > Excellent, and thanks - that was exactly what I was looking for. Only found your mail now, Monday morning mail browsing. :-) I had exactly the same question and we originally built our storage network (17TB, 2 storage bricks and 2 clients, RAID1 was the goal) with backend replication, for the same reason you gave. But after talking to support, we switched to client replication which seems to work great. Taking backends down and starting them up again works flawlessly. The main reason seems to be to avoid the experimental HA module on the client side, to have failover when doing replication on the backend. :-) I'll happily skip experimental modules in production and it wasn't a great problem. To do the "auto self healing" I also have a cronjob to do a find /gluster_volume | xargs stat >/dev/null once every now and then to trigger healing on the volume. If backends go down after each other, we don't want to have missing backup volumes on that Gluster space. :-) And to monitor replication I just "touch" a file from the frontends and monitor the timestamp on the backends. Good luck! /Robin