hi @all,
im using proxmox 3 for a very small company with a dedicated 2 node iscsi storage cluster and a couple of storageless hypervisor nodes.
im not very happy with the current io performance, proxmox 4 is out and so i consider my future options.
i've setup a small 3 node lab cluster with small machines and some old disks to test drbd and ceph on the hypervisor nodes.
both ceph and drbd9 are working fine with an old 500G disk in each node for drbd storge or ceph osd. drbd/ceph network is a dedicated 1G NIC in each node.
i deliberatly used these few disks to have some sort of worst case - one of them has some bad blocks too.
i just wanted to see how performance and reliability is under such worst case scenario.
installation and ceph/drbd deployment both went fine.
migrating an existing ubuntu vm to either storage was ok. running this vm worked too.
some preliminary tests with dd worked 2 times.
the 3rd time i got an io error inside the vm on vda1 and the fs would be remounted ro ...
this happend both with ceph and drbd, restarting the vm would let it run without future errors.
cant say if the dd test data got corrupted in the process.
my questions regarding my lab cluster:
tja...
im using proxmox 3 for a very small company with a dedicated 2 node iscsi storage cluster and a couple of storageless hypervisor nodes.
im not very happy with the current io performance, proxmox 4 is out and so i consider my future options.
i've setup a small 3 node lab cluster with small machines and some old disks to test drbd and ceph on the hypervisor nodes.
both ceph and drbd9 are working fine with an old 500G disk in each node for drbd storge or ceph osd. drbd/ceph network is a dedicated 1G NIC in each node.
i deliberatly used these few disks to have some sort of worst case - one of them has some bad blocks too.
i just wanted to see how performance and reliability is under such worst case scenario.
installation and ceph/drbd deployment both went fine.
migrating an existing ubuntu vm to either storage was ok. running this vm worked too.
some preliminary tests with dd worked 2 times.
the 3rd time i got an io error inside the vm on vda1 and the fs would be remounted ro ...
this happend both with ceph and drbd, restarting the vm would let it run without future errors.
cant say if the dd test data got corrupted in the process.
my questions regarding my lab cluster:
- as far as i understand ceph, the vm should get not io error whatsoever (see http://tracker.ceph.com/issues/8226#note-2) - am i wrong with this ?
- how is io error handling in drbd9 - wouldnt one expect to get BBR of the disk and continue gracefully ?
- are there some config params which would accomplish this desired 'gracefull' behavior ?
- what strategies are recommended dealing with faulty disks ?
- what is the technology of choice for proxmox 4.x++ ? what is it you guys expect to support more/better in the future ?
tja...