Hello! I would like to read your thoughts regarding a new implementation we are about to do at work Our infrastructure is as follows There are 9 servers who will be used for running vm's Network is 2x10G + 4x1G ethernet per server each has 2x 10C xeon processors and 192GB RAM and 2x 300GB 10K Enterprise SAS 6Gbps, Hot Swap HDD There are 3 servers who will be used for storage. Network is 2x10G + 4x1G ethernet per server each has 2x 8c Xeon processors with 128GB RAM 2x 300GB 10000RPM, 2.5", Serial Attached SCSI (SAS), Hot Swap 6x 200GB WesternDigital HGST ultrastar HUSMM1620ASS200 ssd Each server has a jbod box connected to the server through sas interface 12x 4TB The ovirt installation is expected to run 250-200 VM's on maximum load. Expected to run on a big VM there is our mail server which is pretty loaded, has about 50000 registred users and about half of them are pretty active. What I'm currently thinking is : have hosted vm on compute servers storage, as a stand alone non managed volume of about 150GB. but I would like the engine to be stored on more than 3 servers. Any thoughts ? On the vm storage since high availability is a must I will probably set the servers in raid6 with a spare each and then I will do 3 volumes(2 data bricks + 1 arbiter brick on each server) on glusterfs replica 3 arbiter 1. This will give me about 36TB usable per storage server. I'm highly concerned on how glusterfs will perform regarding the email server. Right now email server is on a vm running on KVM and it gets it's storage from an EMC storage. I'm not expecting Gluster to perform as the EMC storage but I would like to know if it can handle it. I'm thinking of dedicating a gluster volume exclusively Is it better to give to the email vm the gluster volume as a disk(s) from ovirt or directly mount it through the vm? ovirt will force the use of gfapi if I set it on the engine. on the other hand if I use this, how sharding will perform. On the other hand, If I mount it as a glusterfs volume directly on the vm how will it perform through fuse? I have 6x200gb ssd's per storage server. Will it do any good to add these as a tiering volume? What should the shard block size be in that case should it be smaller to utilise the ssd's better ? Will performance be better if I use ceph through iscsi with multipath? I read that ceph needs more nodes to perform suitably and that iscsi adds considerable overhead. I'd love to read your thoughts on this.