COE NewsNet – February 2007

Virtualization is generating genuine excitement in the IT community. At a fundamental level, virtualization involves decoupling a workload and its data from the functional details of the physical platform on which it is hosted. This decoupling increases the flexibility with which the workloads can be matched with physical resources, enabling administrators to develop business-driven policies for delivering resources that are appropriate given specific time, cost, and service-level requirements. Virtualization therefore potentially enables IT operations to be performed with far better economies of scale, maximizing the utilization of existing resources by allowing infrastructures to be managed efficiently even as they undergo high rates of growth.

As the adoption of virtualization increases momentum and moves beyond the stage of simple experimentation with the technology, users are becoming aware of certain strategic decisions that they have to confront as they start to deploy virtualization for production workloads. Some of the factors that users have to consider include choosing which workloads are suitable for virtualization; which virtualization technology is most appropriate for a particular workload; how to deal with the impact of virtualization on parts of the infrastructure beyond servers, including storage and networks; and how to adapt management tools and procedures as virtualization is extended from individual servers to entire infrastructures.

The first decision that users generally encounter is which applications to virtualize. On any virtualization platform, applications running within a virtual machine incur a performance overhead. This penalty results from the need to perform a translation or emulation procedure every time a guest operating system running attempts to execute a “privileged” instruction, i.e. a low-level instruction the rights to which can only be granted to the host OS in order to maintain consistency. The actual performance overhead will vary widely depend on the nature of the workload. Among other factors, the more I/O that an application depends on, the higher the overhead will be when running in a virtual machine. Support for virtualization in server hardware help greatly to reduce this overhead, but the maturity of virtualization hardware varies quite a bit by platform -- x86 servers still lag mainframes and RISC platforms in this regard. In any event, users should carefully study the load characteristics of workloads over time before virtualizing them, measuring seasonal demands and differences between dependencies on CPU, memory, and I/O resources.

Much of the recent interest in virtualization on x86 hardware has been driven by VMware, which broke new ground by bringing server virtualization capabilities to industry-standard x86 systems, where before virtualization was largely limited to high-end systems such as mainframes. VMware introduced the ability to run multiple operating systems simultaneously on a single x86 server, which is the fundamental building block of virtualized infrastructures. Following VMware’s early success, a variety of alternative virtualization solutions for x86 systems have now appeared. Many users are particularly interested in Microsoft Virtual Server, which is available for free and has particularly strong support for the Windows operating system. VMware currently has some significant technical advantages over Virtual Server, but Microsoft will ship an improved next-generation virtual machine platform (code-named “Viridian”) after the release of its next server operating system (code-named “Longhorn”), which is due in the second half of 2007.

There is also a lot of interest in Xen, an open-source virtual machine platform that is available from a variety of vendors, including Novell (with its SUSE Linux Enterprise Server 10 distribution of Linux), XenSource, Virtual Iron, and soon Red Hat, which will include Xen support in its upcoming release of Red Hat Enterprise Linux 5. Other virtualization options for x86 systems include Parallels, which specializes in virtual machines for Macintosh systems, and SWsoft Virtuozzo, which implements a variant of virtualization called “virtual servers”, in which only a subset of operating system functions are virtualized, rather than an entire computer system. Finally, the Linux kernel development community is continuing to tinker with a variety of open source virtualization projects, including Kernel-based Virtual Machine (KVM) and User Mode Linux (UML).

Further, higher-end platforms such as UNIX systems and mainframes are not standing still, and benefit from continued innovation with very strong virtualization functions compared to standard x86 systems. Some of these platforms benefit from superior virtualization hardware, which results in lower overhead and higher reliability, and therefore may be more appropriate for virtualizing the most critical workloads. Managers in organizations that have heterogeneous IT infrastructures need to consider a variety of factors in choosing which server platforms to use for virtualization, carefully considering cost, performance, and reliability requirements.

While much of the attention in the industry has been on virtualizing servers, it is also possible to implement virtualization at other levels of IT infrastructure, including networks, storage, and possible even desktops. Storage virtualization may become particularly significant, because some of the more sophisticated server virtualization techniques, such as live migration of virtual machines, often depend on tight integration with virtual storage functions such as Storage Area Networks (SANs) or Network Attached Storage (NAS).

Finally, virtualizing servers and storage is in fact only the first step toward achieving real Total Cost of Ownership (TCO) improvements. Improved server and storage utilization resulting from virtualization does help lower acquisition costs, but those savings may be relatively insignificant compared to the ongoing costs of managing infrastructure needed to host virtualized workloads. Indeed, users are becoming increasingly concerned with determining which management tasks are the most time-consuming and repetitive, and thus potentially benefit the most from the cost savings of automating repetitive tasks, which could be simplified after virtualization is deployed. Some of the management issues that potentially rise to the forefront in virtualized infrastructures include bridging differences between managing virtual machines and physical servers; gaining the ability to rapidly provision virtual servers; instrumenting and monitoring virtual and physical resources simultaneously; integrating virtual management tools with physical hardware management frameworks; implementing distributed resource management, which involves matching workloads with resources regardless of where they are located on the network (and may represent a prelude to deploying full-fledged Grid Computing infrastructures); and correlating virtualization with ancillary initiatives such as blade server deployment, and power and cooling management.

Indeed, over the long term, virtualization may cause a fundamental transformation in the way that IT infrastructures are deployed and operated. As a result, the decisions that users make about deploying virtualization now could have a far-reaching impact. The stakes are high for vendors of IT solutions as well, and their competitive posturing will accelerate as each tries to establish a foothold in a rapidly evolving marketplace. The arrival of mass-market virtualization has truly energized the industry, and it has proven real, practical value for organizations of all sizes. However, users should make every effort to look beyond the hype and marketeering and take the long view as they plot their course through this wave of innovation. The key is to understand the impact that virtualization will have in their own environment, i.e. in the context of their own mix of platforms and operational procedures, rather than in some idealized vision that a particular vendors may be espousing.

Ideas International analysts recently interviewed 50 end users who have deployed virtualization about the platforms they chose for virtualization, the policies they developed to classify workloads for virtualization, and the changes they foresaw in their system management practices as a result of virtualization. A summary of the results of this study will be posted at in mid-February, so stop by the site to pick up a copy then. Also, I will be giving a presentation on the choice of virtualization platforms at the COE 2007 Workshop - Develop & Deploy in Las Vegas on May 3-4.

Contact Tony:

Ideas International, Inc.

800 Westchester Avenue, Suite S620

Rye Brook, NY 10573

(914) 937-4302 ext. 229

tonyi@ideasinternational.com

Visit http://ideasinternational.com for more information.