System Center App Controller and Virtual Machine Manager modules work in concert to control virtualized environments
The System Center 2012 modules that we previously tested — Orchestrator and Configuration Manager — require forklift upgrades. But the modules we tested this time around – App Controller, Virtual Machine Manager and Data Protection Manager — are more graceful and, in some cases, more powerful.
The most interesting combination is the new App Controller coupled with Virtual Machine Manager (VMM) 2012. These two modules are peas in a pod. App Controller deploys VMs into fabrics, which can live in private clouds, plain old hypervised locations or public clouds, especially Windows Azure.
Although VMM doesn’t deliver the depth of control that you can achieve with the tools that Citrix and VMware provide for their own platforms, you can successfully manage a mixed environment of Microsoft Hyper-V, Citrix XenServer and VMware vSphere with the Microsoft management tools.
The final module, Data Protection Manager, is the least exciting, yet a mainstay of all systems everywhere. Data Protection Manager is comprised of a set of monitored backup components for Microsoft applications and servers. While it was boring, we were thrilled to find that one new feature, a bare-metal restore, not only worked the first time, but presented no adrenaline moments at all.
We installed these and the other modules via the Unified Installer; readers are cautioned not to attempt discrete deployment of these modules. Like other System Center 2012 apps, they require Microsoft SQL Server 2008 R2 and SQL Server, which is recommended to be placed on alternate instances.
The most tempting and overarching feature we wanted to test was the ability to have VMM control over Citrix XenServer and VMware vSphere ESXi infrastructures. We have Hyper-V, but also both of Hyper-V’s strongest commercial competitors in the lab and one hefty server instance of all three (See How we did it). We also pestered Microsoft for an Azure account, which the company supplied to us to test cloud control.
We found some Draconian changes had to be made. As an example, we use Network File System (NFS) as a large storage pool hosted on our Dell Compellent SAN. Another NFS pool is located on another server. To use these pools, VMM required us to change them to read/write, which violates a security problem in our workflow. Inasmuch as VMM 2012 essentially takes over workflow (and subsequent controls are administered through App Controller), we moaned a bit and changed the permissions of our pools.
Then, by name, we told VMM where to find our Citrix and ESXi hypervisor metal, and in short order, VMM snacked on all of the live VMs and switched off VM assets within each server type. Each resource type, XenServer and VMware, is then made available. Various tools need to be installed, and there are other considerations to moving VMs around, especially on an inter-fabric basis. A live VM move from one platform to another isn’t possible, but it is possible under confined circumstances from one type of fabric to another of the same hypervisor family.
We tested proof-of-concept transfers of XenServer-XenServer and vSphere-vSphere with success. This doesn’t mean that advanced hypervisor infrastructure family features such as resource matching and advanced storage pool optimization were available — they weren’t. But we had, for simple features, Microsoft’s purported “single pane of glass” control of three different, highly competitive families of hypervisors hosting Windows 2008 R2 server instances.
The concept works — although the top-end features of especially VMware vSphere aren’t available.
VMM 2012 installation can also optionally include a self-service portal that’s accessed via IIS. There’s a layer of management difficulty added by selecting the portal, as many more roles need to be defined and resources configured. The payoff is a Web-accessible VM provisioning method that allows a more appliance-like service role to be offered.
The inner plumbing of the infrastructure associated with what VMM 2012 knows about VMs is largely hidden, allowing for a more appliance-like, application-focused instance deployment. Services can be rendered in a way that allows user self-identification with appliance or VM asset deployment, but here, Active Directory user roles must be controlled through group enrollment with fairly sophisticated Group Policy objects, requiring administrative overhead for the self-service portal to be adequately constrained from abuse.
App Controller then divides up the resources into fabrics, which can be public or private hosting resources. We obtained an Azure account from Microsoft, to test moving VM instances from our private network to an Azure instance, and found that once initially set up (Azure credentials, keys, fabric connections) that VM instance resource movement between on-premises and Azure resources can be accomplished by drag and drop if desired. The downside is that instances spawned within Azure are cheaper, especially SQL Server, than dragging them across the Internet into a new home at Azure.
App Controller can also package groups of VMs as an object for purposes of managing several VM instances. These might be, as an example, several Web servers with a database or service application back end. App Controller actions also can be the crux of PowerShell commandlets (cmdlets) that allow most all functions to be scripted together and, if desired, fed with arguments to allow a basic script, such as one that adds and removes user roles, en masse if desired.
Shell scripts can be reused endlessly, automating frequently used UI clicks. The downside to these and other PowerShell scripts is that while somewhat self-documenting, they’re code that doesn’t have to meet best practices standards, and can become integral to the use of App Controller.
Between the two, it’s possible to do bare-metal deployments of VMs via Windows Deployment Services (WDS), which uses PXE (Preboot eXecution Environment) to essentially hatch VMs on a local network for deployment elsewhere. Procedurally, a machine wakes up, sends out PXE signals to a DHCP server, which in turn, has a preselected image that’s loaded into the machine. This is confined to a local network (and only under IPv4, we’re told) unless a router can be programmed to send the initial PXE traffic to another destination to service the machine’s netbooting request. Once initialized, the server can then, through various means, touch and load other applications or resources to itself.
In all, we rather like modifying virtual hard disks (VHD) then moving them to a PXE boot server (WDS) — but we prefer using NFS in the lab because it services all operating systems for bare metal.
The combination of App Controller 2012 and VMM 2012 is comparatively powerful, and can amalgamate control of resources under System Center 2012 in a very tidy way. We don’t believe that it replaces the secret sauces in the products that it now covers, and the advent of Windows Server 2012 editions and requisite System Center 2012 service packs will change the character of these two seemingly inseparable modules one more time. Much heavy lifting initially can lead to a lot of desirable production control over VM assets and their management costs.
System Center 2012 Data Protection Manager
The updated version of DPM is more focused on Microsoft applications, like SharePoint services, SQL Server and Exchange. Less emphasis is made on using tape within a hierarchical storage scheme, and we readily admit that we’re not using tape drives in deference to online SAN storage in our test lab.
The compliance, regulatory and litigation support requirements that organizations face often lead to online retrieval mandates. DPM does a good job of backing up a number of “behaving” applications, which means Microsoft apps. DPM doesn’t extend well to other operating systems or file systems.
Procedurally, protection groups are assigned, and storage pools allocated. The DPM storage pool can also have custom volumes assigned to it. Microsoft recommends that storage pools be twice the size of projected data that will need “protection.” There are recovery points that can be established, but a limitation imposed of 64 aggregate points per object — not a limitation of application data, just the number of maximum snapshot recovery points. You can have eight scheduled recovery points for each protection group each day, says the system planning guide, which we found to be a very reasonable number.
Microsoft recommends that each DPM server can store up to 9,000 snapshots of data, which can be sourced from clients, servers or server-based applications. The formula for recommended disk space is a variable that is unique to each organization’s data-change rate, and needed for retrieval across the limits of speed of a network architecture. There are suggestions, we found, regarding all of these characteristics of both backup and retrieval/restoration needs, but each organization using DPM will decide what combination of storage, network and frequencies meets its needs for availability and recovery.
Setup of DPM from the Unified Installer is strongly recommended, and, like the other modules of System Center 2012, a link to SQL Server 2008 R2 is needed. Indeed, the first item we wanted to attempt to back up was our SQL Server database. Setting it up took longer than the actual copy, although the databases and tables were at near minimums.
We called upon our Compellent SAN, partitioned partially as an iSCSI target of the SQL Server R2 instance we used in our tests. We used DPM 2012 protection policy formulas to discern that a retention range of five days would take five snapshots. Snapshots can be scheduled for non-busy times, if desired, but we chose the recommended 15-minute incremental synchronizations.
Traffic was minimal in our non-busy test environment. The need for additional DPM servers across an organization to take advantage of local network traffic is up to each organization’s deployment desire, but is also somewhat captive to traffic with SQL Server, and the amount and type of storage connected to DPM servers.
Some organizations will be able to tolerate in-band, network iSCSI traffic, while others will need dedicated SAN infrastructure or really fast links to centralized DPM-connected storage to make their strategies tenable. Microsoft suggests a minimum 512Kbps network bandwidth when using DPM across WAN links, but unless the link is dedicated, we can see it becoming jammed with traffic unless synchronization goals are modest at best.
We backed up Windows 7 VMs and notebooks across our 3Mbps links via a second DPM instance that we brought up in the lab, linking it to our network operations center (NOC) 70 miles away. Each had approximately 21GB of mixed data. Our host in the NOC was an HP DL580 with 16 cores, two for the instance as a VM, connected to another instance hosting SQL Server and other inactive System Center 2012 modules. That, in turn, connected to our Dell/Compellent SAN.
Run individually, the DPM backup of the Windows 7 instances initially took 94 minutes for full backups. Running concurrently, two backups took 109 minutes. Three of them took 126 minutes, concurrently. Finally, adding a fourth full backup and running them concurrently took 177 minutes, and essentially cratered any useful backup in the lab.
You can team together network interface cards to be used with DPM, but as we’re bandwidth-bound, doing so doesn’t help in our mini-WAN although it could help plentifully in healthy Gigabit Ethernet LAN environments, or in DPM areas that service server farms.
Data Protection Manager 2012 lives more autonomously than other System Center 2012 modules, as though its purpose is a necessary burden on the rest of the processes and maybe IT itself. It’s not integrated inside the other modules in a way that says: Oh, don’t forget to do the backups, evil and as boring as it is.
It lacks integration with the rest of the modules in a way that’s different from the interaction between App Controller and VMM, as an example. None of the modules have a workspace that says: Here’s how we back you up, and here are the mindful restoration points, should they be needed. Barring that integration, however, DPM does the grunt work, if for the Microsoft portion of a system’s fabrics. Like the other modules, it requires planning and much thought put forward as to most all of the Windows-based systems processes, and their need for recovery, storage pools, network traffic and pools of VM and process objects that need to be protected.
Bare-metal recovery is supported for Windows 2008-plus editions, but not Windows XP-2003 (we’re still checking on 2003 R2), although system state protection for snapshots on these systems is supported. To enable BMR, we added a test server into a protected server group with a Create New Protection Group Wizard.
The replicant/backed-up volume has extra storage, by default 30GB, and can be increased if needed. The size can be changed later if needed, according to the supplied documentation, but our test was simple: Take a “stock” Windows 2008 R2 server instance, and make it BMR onto actual bare metal. A wizard allowed us to establish which backup to use, and in a few clicks and a cup of coffee, the server was restored. It’s somewhat easier still when using Hyper-V based VMs to restore, as the initial “metal” is already prepped.
Summarizing
What we’ve left out is that most of the activities we tested have a PowerShell version that’s made up of discrete cmdlets coupled to scripts that we could generate if so minded. We eschewed the scripts as we feared losing them or mislabeling them in some way. The power of the script commands begs for better control, perhaps a script generator that also places them into a library with metadata for easy extraction.
System Center 2012 can be deployed minimally, but its strength is in the modules interaction, and this requires much work if it’s a fresh installation. Organizations heavily invested with Microsoft infrastructure (and especially Hyper-V clouds) of medium to large size will find the controls within App Controller and VMM 2012 to be a must-have.
The overall control power of the entire System Center 2012 package, including the Orchestrator and Configuration Manager components (all of which might be backed up with Data Protection Manager), is a formidable combination of components, if at the sacrifice of a tremendous amount of planning, and a disciplining of resources and activities to accommodate these packages. The minimum number of servers would be nine, by our count of the recommendations, but these could be housed in a single hefty VM server (with requisite DPM instances). It’s a truckload of control that specializes in Windows, but plays well with others.
Tom Henderson is principal researcher for ExtremeLabs, of Bloomington, Ind. He can be reached at kitchen-sink@extremelabs.com.
