Virtual servers, storage, and failover

[iSteve]

Hello,

 

We are looking to replace our antique file storage and I am trying to research a better way of doing things that will yield better performance and more resilience than our 4 file servers with local storage. Without talking specifics products/models, I am looking to clarify the building blocks of my potential set up -

 

I am considering a virtual machine host, running two VMs via ESXi. I have read that I can turn on 'fault tolerance' to have these mirrored to another host so that if the host or VM fails, the second machine will take over. Is this correct? We are running some VMs on ESXi already, but have not delved in to fault tolerance.

 

These two VMs will be file servers, pointing to 2 volumes of iSCSI storage on a separate LAN. If I want my storage to be fault tolerant in the same manner, then I will need a second iSCSI storage 'box', but what is the preferred method to sync these up - should I use Windows Failover Clustering (again, new to me) - and if so does this require a separate server to manage the cluster of my two iSCSI storage boxes?

 

My aim is to have a solution where both the VMs and the storage will remain online and in sync.

[Asgard]

Cluster shared volume? So the VMs sit on the shared storage , I'll elaborate based on our setup.

 

We have 4 servers running Hyper-v (on SVR2012 R2) each with a dual 10gb NIC in - one to the LAN and one to a dedicated switch for iSCSI.

 

An iSCSI target ( SAN ) we have a 18tb volume that all of our servers see and use - and all of our VMs are hosted on it, with config stored on whatever server we set it up to use.

 

When we reboot (or force power off a host for DR testing) , actually this can better explain it than me rambling on

 

http://www.msserverpro.com/implementing-windows-server-2012-hyper-v-failover-clustering/

[dhicks]

These two VMs will be file servers, pointing to 2 volumes of iSCSI storage on a separate LAN. If I want my storage to be fault tolerant in the same manner, then I will need a second iSCSI storage 'box', but what is the preferred method to sync these up - should I use Windows Failover Clustering (again, new to me) - and if so does this require a separate server to manage the cluster of my two iSCSI storage boxes?

 

If you buy two iSCSI storage devices of some kind, or set a couple up yourself using FreeNAS or similar, they will have their own mechanism for sycning between them. We have a setup where we sync directly between two VM hosts, each with local storage - each host gets local-speed storage, plus VM storage gets mirrored to a second server and we don't have to buy any iSCSI devices. This is using DRBD, used as the inter-device syncing mechanism by several iSCISI storage appliances, and with XenServer as the VM system - I'm sure there's a Microsoft and VMWare equivilent.

 

This is proving to be stable, but not the most flexible of setups - you wind up having to mirror a whole XenServer storage repository at a time, and only one VM host at a time can even see that storage repository. It's not quite the slick system I was hoping for, where if one server went down VMs would just keep on running as if nothing had happened. I plan to have a look at Ceph for VM storage instead, with hosts doing both computation (being VM hosts) and storage (being Ceph nodes). This, again, should avoid the need for any centralised storage, and gives us a mechanism for offsite mirroring of data via part of the Ceph cluster being at another location. I think the Windows equivilant of Ceph is Storage Spaces, available in Windows 8 and Server 2012.