Sunday, February 21, 2010

Windows Server R2 8

You can cluster multiple Hyper-V servers (the host machines). To build a Hyper-V cluster, you need two Hyper-V servers with identical hardware configurations, and a separate SCSI-3 compliant storage appliance (SAN/NAS Storage, iSCSI is a good option). On the storage you need two volumes, one volume for quorum and one for storage.
When one machine needs to go down for maintenance, the VM can be migrated to the other live machine. This means that the RAM of the machine can be moved over, without shutting it down.

There are two methods:
1) Quick Migration
The VM is first hibernated, moved (in terms of the service) to the other node, then brought back up. This was already available in Server 2008 R1. This downtime may not be good for the SLA.

2) Live Migration
The VM's RAM is copied over to the other node while the original machine is still running. There is no downtime. We don't have to know the details, it's all done automatically.

Typically to set up a Hyper-V cluster, you'll need a minimum of $20,000 worth of hardware. However, you can "hack" the server so you don't need to use a compliant iSCSI storage and two similar servers. However we do need a NAS storage appliance.

A "heartbeat" is like "hello" messages which components in a failover cluster uses to make sure that everyone is alive. This "heartbeat" needs to be sent out from a NIC dedicated to it.

Failover Clustering needs to be installed on both Hyper-V hosts. After installation, you would be able to run the Failover Cluster Manager.

To attach an iSCSI storage, we need to use the iSCSI initiator. The one in R2 has a Quickconnect tool which was missing in 2008.

After connecting to the disks, go to Volumes and Devices and click on Auto Configure to mount the volumes.

Note that we need to create iSCSI targets from the NAS before Windows will detect it. After adding them from the iSCSI, we would be able to access them from Disk Management. Failover clustering wizard requires that the disks be OFFLINE before it would proceed. The iSCSI initiator needs to be run from both systems (all the participating systems).

In the Failover Cluster Manager, there is a "Validate a Configuration Wizard" which would check if our servers are compatible.

If your storage doesn't support SCSI-3 persistent reservations, you will not pass the validation test. However, you can get around this.

Here's how we work around the problem. We first disconnect the two disks from the iSCSI initiator. After this, it should disappear from Disk Management. After this, we'll run the validation again. We'll be able to proceed with some warnings.

After this, we'll run the Create Cluster Wizard to create the actual cluster.

We'll then need to give the cluster an IP address. This is a virtual IP similar to HSRP. This is not load balancing in a sense that it does not distribute the requests.

There are four types of quorum configurations:
1) Node Majority
This is not recommended as it cannot sustain a failure of any of the nodes.

2) Node and Disk Majority
If the quorum disk is online, it can sustain failure of 1 node.

3) Node and File Share Majority
If the quorum disk is online, it can sustain failure of 1 node.

4) No Majority
Can sustain the failures of all nodes except 1.

To do this hack, we'll need to choose Node and File Share Majority, which uses a share location as a quorum disk. The quorum disk may also be referred to as a witness disk. We can get to the configuration by right-clicking the cluster from the Failover Cluster Manager.

In Server 2008 R2, we can create Hyper-V machines directly from the Failover Cluster Manager. We do this by going through the New Virtual Machine Wizard. This time, however, we create a disk in our system disk. This allows to proceed, but is not really functional in terms of high availability.

No comments :

Post a Comment