Powered by System Center
Archive for February, 2010
DPM 2010 is nu in Release Candidate. Ik gebruik dit voor backup van Exchange 2010 en dit gaat erg goed. Bij wijze van test gebruik ik het ook om VM’s op Hyper-V R2 te backuppen, en ook dat gaat erg goed.
Microsoft heeft er een whitepaper over geschreven en die vind je hier:
Binnenkort zal ik hier meer over schrijven.
it has been too quite for some time, but I have been extremely busy I’m afraid. But, I’ve written an article on the Simple Talk website about Hyper-V R2 Live Migration…
Windows Server 2008 Hyper-V offers a high availability solution by using Windows Server 2008 Failover clustering. The Virtual Machine is implemented as a cluster resource and when a host node fails, the Virtual Machine resource fails over to the other node. Like all other Windows Server 2008 Failover solutions, the resource is brought offline before it actually fails over. This results in a relatively small period of downtime which is unacceptable in certain environments.
Windows Server 2008 R2 Hyper-V offers a Failover clustering solution without any downtime for the Virtual Machine. This solution uses a new feature called “Cluster Shared Volume” or CSV and is called “Live Migration”. In this article I’ll explain what the CSV solution is and how it works.
Windows Failover Clustering
When creating a Highly Available Hyper-V environment, a Failover cluster needs to be created. A Failover cluster is a logical server consisting of two or more Windows Server 2008 servers. Windows Server 2008 supports a maximum of 16 servers in a Failover cluster.
These servers are called Cluster Nodes. All Cluster Nodes in the Failover cluster are connected to a shared storage solution where the data is stored. The Virtual Machines running under Hyper-V that need to be highly available are configured as a resource in the cluster.
Note. All servers in the cluster must run the same Operating System, all nodes must either be Enterprise or Datacenter Edition and for Live Migration all servers must run the same processor family (even the same processor stepping).
In a Windows Server 2008 Failover cluster, only one Cluster Node can be “owner” of the disk resource and only the owner of the disk resource can access the files (including the Hyper-V VHD files). Therefore, if a Virtual Machine needs to be failed-over to another Cluster Node, the complete disk resource needs to be brought down, moved to the other Cluster Node and be brought online again. The Virtual Machine is in a “saved state” during a cluster Failover. To bring the Virtual Machine in a saved state a certain amount of time is needed. For a Virtual Machine with only 256MB of memory this is only a short time, but a Virtual Machine with 8GB or 16GB can need up to minutes to be brought into a saved state. Starting the Virtual Machine on the new node will take the same amount of time. Needless to say, failing over a Virtual Machine can cause a significant downtime.
The complete article can be found here: http://www.simple-talk.com/sysadmin/virtualization/hyper-v-r2-live-migration/
VMware is targeting MS for not using Hyper-V for their virtualization platform. Well in fact they are not using Hyper-V current version but Hyper-V vnext (possibly even vnext R2) version. Hyper-V R2 already benefited from the experiences with Azure Hypervisor and the upcoming Hyper-V or even the next service pack will benefit in an identical fashion.
Design Principles Behind The Windows Azure Hypervisor
By Hoi Vo, Director on the Azure team
We are frequently asked about the Windows Azure Hypervisor, and whether or not the code will be made available to customers as a product they could run in their own datacenters. We built the Windows Azure Hypervisor with three principles:
- Efficient: push work to hardware as much as possible. Any percentage gain once multiplied to tens of thousands of machines will be very significant for us. Consequently we can bet on new processor features to save CPU cycles for the hosted application.
- Small footprint: any features not applicable to our specific cloud scenarios are removed. This guarantees that we do not have to worry about updating or fixing unnecessary code, meaning less churning or required reboots for the host. All critical code paths are also highly optimized for our Windows Azure scenarios.
- Tight integration: The Windows Azure Hypervisor is tightly optimized with the Windows Azure kernel. This is required to achieve the level of scalability and performance we want for our stack.
Much of the development for the Windows Azure Hypervisor would only work in our environment, taking advantage of our specific homogenous data center environment. Some of the innovations would be useful to customers with a different data center design and will be incorporated in future releases of Hyper-V (e.g. Second-Level Address Translation will be available in Hyper-V v2.0).
A hotfix rollup is now available for System Center VMM2008 R2 which is part of the February 9 update.
List of issues that are fixed
Consider the following scenario:
- A user is a member of the Self-Service User role.
- A new virtual machine is created, and the user is assigned the owner of the virtual machine.
In this scenario, the user cannot connect to the virtual machine by using the Self-Service Portal.
Consider the following scenario:
- A highly available virtual machine is offline.
- Quick Storage Migration is used to move the offline virtual machine to another storage location that is in the cluster.
In this scenario, the virtual machine does not migrate to another node that is in the cluster by using Quick Migration or Live Migration after Quick Storage Migration is used.
VMware virtual machines are not listed in the Virtual Machine Manager console if there are duplicate custom named tags in VMware Virtual Center.
Consider the following scenario:
- A new virtual machine is created, and the owner of the virtual machine is a member of the Self-Service User role.
- While the create virtual machine job is running, a user is added or removed from the Self-Service User role.
In this scenario, the user is not added or removed from the Self-Service User role because the create virtual machine job is running.
This rollup also includes the resolutions that are documented in the following article for a Virtual Machine Manager 2008 R2 hotfix rollup package:
976244 (http://support.microsoft.com/kb/976244/ ) Description of the System Center Virtual Machine Manager 2008 R2 hotfix rollup package: November 10, 2009
After the hotfix rollup package is installed on the Virtual Machine Manager server, update the VMM agent on the Hyper-V and Virtual Server hosts.
Install at least these two hotfixes on Hyper-V R2 servers:
KB975354 A Hyper-V rollup package dealing with parallel backups on same cluster shared volumes and VM’s hosted by different servers; Avoiding data truncation as a result of simultaneous backup of VM from host and from within virtual machine; properly restoring a VM with snapshots to another location
KB975921 You may be unable to perform certain disk-related operations after an exception when a hardware provider tries to create a snapshot in Windows Server 2008 R2 or Win 7.
Although not specifically related to Hyper-V and DPM2010, I also install hotfix KB974909 which solves loss of the network connection within a virtual machine with heavy outgoing network traffic and many concurrent network connections. This could easily be the case when a VM level backup is being conducted.
There is also a list of tested hardware VSS providers:
I may have a chance to test the HP LeftHand VSS hardware provider, which is not yet on this list. The HP EVA 4×00, 6×00 and 8×00 is on the list by the way.
A comprehensive list of Windows updates and hotfixes for Hyper-V and Hyper-V R2 can be found here:
If you need to update your Hyper-V servers, you might just as well include this recent security update:
Microsoft Security Bulletin MS10-010 – Important
Vulnerability in Windows Server 2008 Hyper-V Could Allow Denial of Service (977894)
When you try to protect VM’s on a Hyper-V R2 clusters with DPM2010 RC you now get a warning if the required hotfixes are not installed.
This is very helpful!
The release candidate of DPM2010 has just been released.
The download location is:
The RC features support backup of machines within your intranet, which includes:
- Workgroup machines
- Machines in untrusted domains within your intranet
Please note that in this release, this feature has been built for backing up machines within environment and DPM2010 RC does not support backup of machines outside your intranet, which includes:
- Machines in DMZ network
- Machines in untrusted domains on customers’ sites connected through VPN
In a subsequent release (not sure if this is RTM), support for extranet scenario’s might be added.
Support for Hyper-V R2 clusters and cluster shared volumes (CSV) was already available in the beta.
Important to know is you can upgrade from beta to RC to RTM, or directly from DPM2007 to RC or RTM or any other combination.
In our practice we have deployed quite a number of beta versions of DPM2010 and we are quite happy with the result. I am glad we are getting closer to RTM which is expected to be on schedule for mid April 2010.
DPM2010 RC support can be found here: http://www.microsoft.com/communities/newsgroups/list/en-us/default.aspx?dg=microsoft.public.dataprotectionmanager
In my November blog I talked about moving virtual machines to the cloud. More evidence of an upcoming Infrastructure as a Service (IaaS) can be found on MSDN, which describes virtual machine sizing, ranging from 1-core (Small) to 8-core (ExtraLarge) virtual machines with memory up to 15GB of memory and 2TB of disk space.
Another service described is the ability to mount Windows Azure drives, which act as a local NTFS drive, mounted on the server’s file system, accessible to code running in a role. This was previously referred to as an X-drive. Mounting a so called CloudDrive requires Windows Azure Guest OS 1.1 (release 201001-01).
Will we really be able to migrate a virtual machine from our private cloud to the Azure public cloud? I don’t think we have to wait much longer before hybrid clouds become a reality.
Will we be able to move a VMware VM to Azure? I think you can answer that for yourself.
HPStorageGuy tweeted a very interesting reference to an article by Mike Moroze from HP LeftHand about tighter integration with VSS.
A new option in HP LeftHand P4000 iSCSI storage systems, which we use a lot at Nobel in Hyper-V R2 clusters, now fully embraces the Volume Shadowcopy Services framework.
Currently the preferred method for backup up fair quantities of Hyper-V virtual machines is using the VDS hardware provider of the SAN vendor. This would in turn trigger the storage regular hardware snapshot method. Being a 100% VSS solution, I expect DPM2010 to benefit a great deal from this new LeftHand option to off VSS for creating hardware based snapshots. It would alleviate the Hyper-V host to a great extent because IO Redirection can be kept very short for the CSV LUNs on which the VM’s reside which are being snapshotted.
Because there is no similar VMware integration (yet, if at all via VSS), the VSS option for LeftHand is not yet the default. It will be default for most of our customers (grin).
For a single snapshot:
For a schedule to snapshot a volume: