At Fri, 2 Sep 2011 14:23:53 +0200, Valerio Pachera wrote: > > 2011/9/1 Dan Luedtke <maildanrl at googlemail.com>: > > Do you use dedicated NIC for syncing? > > No and I was wondering the way to achive that. > I guest by the parameter "bindnetaddr" in /etc/corosync/corosync.conf right? > Using the address of a nice that is not the one I bridge with qemu-kvm. > > I was also wondering if this scenario is possible: > sda6 /mnt/sheep_data1 -> sync by eth0 > sdb6 /mnt/sheep_data2 -> sync by eth1 > > This way one disk is served by one nic. > Do you think it's possible? There is no way to use multiple NICs now; all the sheep daemons use only one NIC which is specified in corosync.conf, and data synchronization is done through the same NIC. I think it is a good idea to support specifying interfaces used for normal I/Os and object recovery. For example: $ sheep /store_1 --nic io:eth0 --nic sync:eth1 $ sheep /store_2 --nic io:eth2 --nic sync:eth3 > > PS: > I think it's not the best Idea to have more than one sheepdog > directory per node, becaus if the node goes down, much more data goes > down too! But even if you use RAID0, the capacity of the node will increase and Sheepdog should allocate much more data to the node. > A good shenario for me is 3 nodes with 2 disks each that form raid 0 > md device and 2 gigbit nics. > This way the sync could use the full speed of a dedicated nic (1Gbit ~ > 125MB/s) without full fill the disk I/O capacity. > A "normal" sata disk has 60-80 MB/s write speed, a raid 0 is double, > so close to or more than max theorical nic speed. But if you run two sheeps daemon, the I/Os will be distributed over two disks. If we can specify a nic for each daemon, I think there is no big difference between using RAID0 and running multiple sheeps. Thanks, Kazutaka > > node1 (sda/sdb) -> eth0 sync > -> eth1 bridge for kvm guests > -- > sheepdog mailing list > sheepdog at lists.wpkg.org > http://lists.wpkg.org/mailman/listinfo/sheepdog |