On the vendor side, virtualization providers are being forced to bake full production features into their offerings to support whole new functionality sets, including but not limited to:
>> Automatic restart of VMs to support nonstop and fail-safe operations
>> Migration of live workloads between VMs across hosts
>> Migration of workloads across storage devices, including inexpensive server-attached gear
>> Intelligent scheduling of resources based on workload-placement rules
Reasons IT might consider a multihypervisor approach include workloads of varying levels of criticality and sensitivity, test vs. production, collaborative vs. transactional, B2C vs. B2E, not to mention servers vs. desktops vs. storage. Oh, and there's also pricing, licensing, and bundling--an area where some interesting vendor dynamics are at play. Microsoft is upping the ante in the hypervisor space, betting on tight coupling to gain market share; Hyper-V is an integral part of Windows Server and will be free with Windows Server 8 when it debuts next year. Meanwhile, Citrix and VMware are creating "comprehensive cloud OSes" on top of their virtualization platforms, via such offerings as VMware's vSphere 5 with vCenter, vShield, and vCloud plus vFabric and Cloud Foundry (free) as platforms for application development, and Citrix's CloudBridge, CloudPortal, and CloudGateway.
Virtualization leaders are also trying to leverage strengths in one area to grab turf in others, as consolidation of servers and storage picks up, and as control over desktops/handhelds, and related application development becomes a greater priority for IT. For example, Citrix is betting that its position in desktop virtualization, thanks to XenDesktop, can help grow its market share on the server side. To that end, it's offering a free edition of XenServer as an incentive. VMware is trying to leverage its leadership position in server virtualization by strengthening ESX with vSphere 5 (a type 1 hypervisor), and by positioning its View (a type 2 hypervisor) as a more flexible offering for desktop virtualization. In comparison, Citrix's XenClient (part of XenDesktop) is a type 1 hypervisor. Why does that distinction matter? Type 1 hypervisors, like VMware ESX, Hyper-V, or IBM's CP/CMS, are not as close to the bare metal as commonly believed--they have micro-kernel operating systems in them to manage hardware. Nevertheless, type 1 hypervisors have an edge on the performance scale and are superior for workloads involving server virtualization, including hosting virtual desktops. Type 2 hypervisors, like Red Hat's Linux KVM, VMware View, Microsoft's Virtual View and Virtual Server, and Virtual Bridges' Verde, on the other hand, score better on the flexibility scale and may be better suited for workloads involving desktop virtualization and server virtualization for a pool of heterogeneous hardware.
Open source virtualization vendor Red Hat is also getting into the type 2 market with its acquisition of Qumranet, which brings a kernel-based Linux system. Its KVM (type 2) gets support from the Open Virtualization Alliance, whose members include IBM, Hewlett-Packard, and Intel, and is promoted for both servers and virtual desktops.
Coping With Heterogeneity
All this bundling and extending may make for attractive deals, but beware: Introduce more moving parts into the very foundation of the IT infrastructure and you run the risk of sprawl getting completely out of control. VMs being spawned willy nilly negates two main benefits of virtualization-- consolidation and optimization. If you can make a strong case for multiple hypervisors, budget for a hypervisor-neutral management console that has support for unified security, backup, disaster recovery, and load balancing and related migration of VMs; we'll discuss these more in a moment.
As things stand today, there are no easy answers to this fast-evolving problem of multiplicity. But you can make some architectural decisions that will help. First, optimizing business workloads on the most appropriate hypervisor will be critical. While container technologies, such as OpenVZ and LXC for Linux and Parallel's Virtuozzo for both Windows and Linux, which virtualize operating systems rather than hardware, could be tempting, this is cutting-edge technology that may be more appropriate for homogeneous hosted environments and cloud service providers as opposed to enterprise IT. Caveat emptor.
Question whether workload specialization, with the attendant support and training, is really justified. That means detecting the true motivation behind the quest for a multihypervisor setup: Is it just cost, or are there workloads that are unsupported and/or ill-supported with the current hypervisor strategy?
Watch the hypervisor providers; they'll continue to jockey for advantage by either making their systems easier to manage via a single set of tools or by playing the open (and in some cases, the open source) hand. VMware's current position is that its management tools will work only with its own hypervisors, while Microsoft (System Center) and Citrix (XenCenter and CloudStack for open source environments) claim support for each other's hypervisors. System Center 12 also boasts broader hypervisor neutrality, and Red Hat's Network Satellite is an option for an open source, multihypervisor environment.
Claiming even broader multihypervisor management capabilities are pure-play systems management tools, such as HP's new Cloud System, BMC's Cloud Lifecycle Management, CA's 3Tera AppLogic, Dell's Virtual Integrated System and VIS Advanced Infrastructure Manager, and IBM's Tivoli suite for service and virtualization management. For large shops that need this reach, these systems are worth the money now.
We are in the early stages of a disruption that needs to be understood and managed, not feared. As complex as heterogeneous hypervisor environments are, the business drivers can be compelling. As we discussed, lower costs and support for additional workloads will be more and more in reach, as the playing field keeps getting more and more level. Keys to success include:
>> Establish alignment between workloads and hypervisors, and follow up with a technical proof-of-concept that includes all the use cases related to centralized management and virtualization security.
>> Perform a detailed TCO calculation that goes beyond the lure of initial capex and licensing arguments. Account for migration from the current hypervisor and management costs, which could be significant.
>> Think beyond infrastructure workloads when you embark on this journey. Infrastructure management is table stakes at this point. The debate must include middleware and applications as well.
Sreedhar Kajeepeta is global VP and CTO of technology consulting for GBS at CSC. He is based in Farmington Hills, Mich., and can be reached at firstname.lastname@example.org.