Next in this series is a TechEd Europe 2009 presentation by Asim Mitra, a senior program manager from Microsoft, on data protection and recovery of Hyper-V and Hyper-V R2 workloads with Data Protection Manager 2010.

DPM2010 beta has been publicly available for a few weeks. In my own lab DPM2010 is spinning happily, protecting a Hyper-V R2 cluster with Exchange Server 2010 in a VM on cluster shared volumes (CSV) and several Windows 7 laptops and PC’s. It is quite nice to be in control as a user and determine what to protect from my laptop disks and to be able to restore something whenever I want to. But protecting my production mail server on CSV was really urgent, so I upgraded my DPM2007 SP1 server in-place to DPM2010. It even took care of upgrading the SQL Server 2005 database to SQL Server 2008 without hassle.

Before I went to this session, I had already decided that DPM2010 beta is already more convenient and complete that its predecessor.

The number of supported data sources has grown with DPM2010 and several of them were co-authored with the application developer for a smooth protection and recovery solution. SAP, MS Dynamics re examples of these joint efforts. New data sources are Exchange Server 2010, SharePoint 2010, Hyper-V R2. I probably have missed a few.

In DPM2007 SP1 host-level backup of Hyper-V on Windows Server 2008 and Hyper-V clusters was introduced including Quick Migration.

In DPM2010 this support is extended to Hyper-V R2 with Cluster Shared Volumes (CSV) and is now able to protect virtual machines that are live migrating between hosts. Not many users realized that DPM2007 SP1 was only able to restore VM’s to its original Hyper-V host. Microsoft provided a script to work around that. Now DPM2010 can do Alternate Host Recovery and even Item Level Recovery. This last feature is really unique to DPM!

The question has always been: Should I protect from host or guest?

• I want to selectively backup individual data objects like databases & files
• I want to backup each virtual machine as a single object for protection

The following guidelines can be given to answer those questions:

Host level

• Protect or recover the whole virtual machine
• Protect non-Windows servers
• No granularity of backup
• “Bare Metal Recovery” and “Item Level Recovery” of every VM
• Requires single DPM license on host (protecting all guests)

Guest level

• Protect or recover data specifically
  • �
    • SQL Server
    • Exchange
    • SharePoint
    • Files
• No different than protecting physical server
• Requires a DPM license per guest (VM)

Whole Node Protection

• This includes the host OS and all VM’s with host level protection
• Requires same single DPM licence on host as in host level protection

Protecting both the parent and the guest is a very cleverly designed cooperation between Volume Shadow Copy writers and requestors. In the example of a Hyper-V R2 server with multiple guests this looks like this:

The Hyper-V VSS writer interacts with a requestor service which is actually the Hyper-V VSS integration component in each VM.

Because there is some time difference between the VSS snapshot of the host and in the guest, there is a potential for data corruption if this wouldn’t be handled correctly. So DPM2010 takes a post-snapshot step to fix the data.

In this post-snapshot step the VHD in the guest is mounted and the amount of changed bytes or blocks are synchronized between the host VSS snapshot and the client VSS snapshot. It sounds easy and it is easy!

Protecting a Live Migrating VM

A more challenging task is the proper protection of a guest which is in the process of migrating between Hyper-V R2 hosts. DPM2010 is able to handle this as well. It works like this:

1. DPM2010 performs an incremental backup of VM from cluster node A
2. The VM then migrates for instance to cluster node C
3. DPM2010 automatically performs the next incremental backup of the Live Migrated VM from cluster node C

DPM has to be aware of the new shared storage architecture of Hyper-V R2. CSV is implemented as a filter driver and sits directly on top of NTFS. DPM can only make a new incremental backup if it knows how to handle the underlying disk architecture. In a Hyper-V R2 cluster all nodes in the cluster can read from a disk in the CSV pool. They can also write to the disk with the VHD on it. Only the so-called coordinator node has full access to the metadata of the underlying physical disk. This coordinator node is not static and can move between cluster nodes, or else it would be a single point of failure. DPM is clever enough to failover the coordinator role to the node in the cluster that needs to take an incremental backup. So if node A owns the disk, and the VM is moved to node B, the DPM agent moves the CSV disk also to node B. This effectively switches the node from Indirect I/O to Direct I/O which makes it possible to make a local incremental backup of the VHD’s from the correct cluster node. Solved that!

There are several Hyper-V recovery options with DPM2010:

1. Restore VM back to original host or cluster
2. Restore VM to a different host or cluster
3. Item Level Recovery (ILR) to file share

This ILR functionality requires that the Hyper-V R2 role is enabled on the DPM2010 server. Hyper-V will not have to do anything else but attach the VHD. This sounds odd since R2 can do this natively without the Hyper-V role enabled.

Planning the deployment

Normally when I plan for DPM storage I roughly calculate 300% the amount of used production disk to be used for the DPM Storage Pool. This is often enough for a retention time of about 14 days on disk. The applied schedule is not relevant for this calculation, because it doesn’t matter whether you synch the data once a day or once an hour. The amount changes per day remain the same. It becomes a different matter if you also want to protect complete guests. Some data would have to be protected multiple times. With ILR this problem is largely eliminated. But to help planning the deployment, the DPM Team offers a Storage Requirements Calculator For Hyper-V Workloads (which is currently in development). We could download a pre-release from the TechEd website.

Why is DPM a suitable data protection product for Hyper-V?

• Incremental backups only – full once (first replica)
• No more backup window – online backups
• Application consistency via VSS
• Protect Live Migration VM’s in CSV clusters
• Protects whole VM and recovers individual items
• Auto protects new VM’s
• Routine backups (nightly or more frequently)

As part of Microsoft System Center, Microsoft claims that:

• It is the best product for protecting Windows file and application servers
• It is built for Microsoft Virtualization environments
• It is designed for Windows Clients
• It has Enterprise-Ready scalability and reliablity