On 07/22/2014 06:15 PM, Itamar Heim wrote:
On 07/16/2014 06:46 PM, Demeter Tibor wrote:
Hi,
We have a production environment with KVM+centos6 and we want to switch to ovirt. At this moment we have 12 VM on three independent server. This VMs uses the local disks of servers, we don't have a central storage.
currently we have for ovirt
- Two Dell R710 with 128 gigs of ram. - A third dell server for ovirt-engine. - four 1 gb/sec NICs/ server. - Smart GB switch
we would like make an ovirt environment with
- clusterized, redundant filesystem, data loss protection - If a host goes to down the VMs could made a restart on the remain host - LACP/bonding (mode 6) for fast I/O beetwen gluster hosts, we don't have 10Gbe nics - 8 TB of disk capacity for VMs
we don't want:
- using hw raid on servers, because we need free disk tray for more capacity
My questions.
- Which glusterfs method is the best for us for performance?
vijay?
A replicated gluster volume does help for availability and performance. Please ensure that you optimize the gluster volume for virtualization by applying the performance settings in virt profile [1].
- Can I make a real "performance" disk i/o by 4-4 NICs ? Or I need 10 Gbe nic for this? - How much disk need for good redundancy/performance? 4/server or 2/server ? - What will the weak point our project?
make sure to use gluster with replica 3, which needs 3 servers? maybe use hosted engine to remove need for dedicated server for it?
Using replica 3 volumes would provide better protection for split-brains. Thanks, Vijay [1] https://github.com/gluster/glusterfs/blob/master/extras/group-virt.example