IMHO, host level raid means software raid in one local node. In order to support live migration, we should clone the software raid in the destination node and make sure all the data are flushed into the external storage device before migration starts. On 2012-7-29 17:08, Johan Kragsterman wrote: -- Shu Ming <shuming(a)linux.vnet.ibm.com&gt; IBM China Systems and Technology Laboratory

Hi! Dan, it seems you know something about storage, so I adress you: I find the oVirt storage framework a bit odd. It has it advantages, but also a lot of disadvantiages, it seems. I did earlier, in this thread, ask about storage setups you want to use in enterprise environments. And I like to continue that discussion. If you like to replicate on SAN level between, let's say two different LUN's, that should contain the same informnation for a host, or for a VM, then you end up in trouble if you use the oVirt storage system. Because you need to put the two LUN's from the different datacenters together, to become ONE unit, otherwise you don't get the right uuid for live migration. Isn't right? As you wrote earlier, this might be solved with gluster, but is it so, that oVirt is going to get enterprise features only in pair with gluster? I don't believe that is a good idea for the adoption of oVirt. oVirt must have a good support for enterprise features that is used today, like NPIV. I have googled around oVirt and NPIV and didn't come up with much, you can tell me something about it? I belive it would be good if oVirt supported other storage system types than the one system it supports today, that all storage should be imported and handled by LVM and oVirt engine. Most enterprise storage today is fibre channel and NPIV based. You boot most the VM's through NPIV LUN's, you add more space through thin provisioning or more NPIV LUN's. And this is going to continue, even if the Fc world move to FCoE. So perhaps you can tell me a little bit about the storage strategy that oVirt dev team has? Rgrds Johan -----Dan Yasny <dyasny(a)redhat.com&gt; skrev: ----- Till: Johan Kragsterman <johan.kragsterman(a)capvert.se&gt; Från: Dan Yasny <dyasny(a)redhat.com&gt; Datum: 2012.07.29 12:55 Kopia: users(a)ovirt.org Ärende: Re: [Users] host raiding two LUN from different external storage You can use replicated at the SAN level storage between different DCs, for DR. You can also use Gluster for the same purpose. Setting RAID1 between two geographically separate hosts is asking for trouble. ----- Original Message ----- -- Regards, Dan Yasny Red Hat Israel +972 9769 2280

----- Original Message ----- If you are using SAN level replication, it is not an active/active replication. The second replica is in standby, and if the first LUN fails, you can substitute with the replica (of course you can have corruption that gets replicated and then the storage is gone. RAID is never a substitution for backups). With oVirt, if you have an offsite LUN replica, for DR reasons, you can take that LUN and attach to an offsite replica of the oVirt setup (engine plus hosts) - and the original storage domain will be right there for you to use. Once the DR setup is operational, VMs can live migrate within it. Gluster has the replication feature built in. So do most enterprise grade storage solutions. If your SAN supports LUN replication, this feature can be used with oVirt or any other system that keeps data on the LUN, no special magic there. NPIV is completely out of context here. Generally speaking, Linux supports NPIV (see http://fedoraproject.org/wiki/Features/VirtStorageManagement) so if you handle the actual commands, and oVirt hosts see the LUNs in dev/mapper it will work. In very broad terms of course, oVirt does not care how device-mapper gets to see the LUNs, if they are visible, normally, they can be used. If you are talking about incorporating NPIV settings in oVirt engine host management, then it's a feature that needs to be spec'd out in the oVirt wiki first, and since it's an open project, anyone willing to actually develop it is welcome to contribute. You mean clustered filesystems? We already support those, look at PosixFS. But still, LVM provides a huge advantage over a classical clustered FS - less overhead, simple management using standard Linux commands, amazing scalability (what other virtualization solution scales to hundreds of hosts out of the box?) Actually, I haven't seen NPIV used too much, normally, when you have an FC infrastructure in place, purchasing an extra port/HBA is not an issue, and it's so much easier to manage without virtualizing - plug your cables in, set up zones, and you're good to go. So I don't agree about "most", but as I said before, if you want to use NPIV, you can, oVirt will simply not manage it for you, at least not in the current version. Everything is in the wiki - read the storage related feature pages. And if you want to pitch in, you are always welcome to -- Regards, Dan Yasny Red Hat Israel +972 9769 2280