Conceptually, the idea of virtualization is straight forward – computer devices such as servers, networks, personal computers and disk storage are created via software that presents itself as physical hardware. Virtualization concepts also extend to the utilization of resources external to the local computer desktop for processing, packaging, managing and creating utopian operational efficien cy of applications as if everything was currently installed and integrated to the local computer desktop.

For many, virtualization is the Holy Grail they have been searching for as an answer to space constraints and to reduce the recurring costs of power and hardware maintenance agreements while providing high availability, disaster recovery and portability among the many other operational benefits. However, incorrect design and implementation of virtualization can lead to higher management costs, business continuity oversights and operational performance lag. This document will provide a template for design, analytics and planning considerations for use of common virtualization concepts.

 

Definitions

Server Virtualization	The utilization of multiple independently installed operating systems (“virtual machines (VMs)” or “guests”) on the same physical hardware (a “host”). This is accomplished either by installing VMs on top of a host operating system with virtualization software, or by installing VMs on top of a thin layer of virtualization software called the “hypervisor” which has direct access to the hardware and may provide greater efficiency. The VMs may be the same or different operating systems, and do not “know” that they are sharing the same hardware with each other.
Desktop Virtualization	Virtualization of desktops has taken on two flavors:
	Blade Desktops. Pioneered by ClearCube Technologies and IBM, this platform puts the guts of a PC onto a blade (much like a server blade) in the datacenter. End users connect over thin client devices.
	Virtual Machines. Serving up desktops with the same technology as server virtualization (again, typically accessed via thin client devices).
Network Virtualization	A key component of server virtualization technologies is the ability to use and create virtual switches and virtual networks within the same physical server. Benefits include the ability to “hide” certain virtual servers from others, useful to isolate the testing from the production environment, and the ability to extend different IP VLANs onto the same host.
Storage Virtualization	A logical abstraction layer that can pool similar or varied physical storage devices into a set of storage volumes for a host system. In addition to storage device pooling, storage virtualization offers many advanced contingency measures such as mirroring and volume snapshots; over-provisioning of storage space and the ability to migrate data from one volume location to another without downtime.
Presentation Virtualization	The utilization of resources of a remote machine to process and run applications while providing the user interface (display, keyboard and mouse) and control of the application to a connecting desktop computer. Common solutions include Windows Terminal Server and Citrix.
Application Virtualization	The packaging of all shared resources an application needs to run, including registry entries and Dynamic Link Libraries (“DLL”) among other operating system enablers, as a self-contained environment for a specific application to run. This way, applications no longer need to compete for shared resources, which is commonly referred to as “DLL Hell.” The application lives on a remote machine (which can be a Terminal Server) and is initially launched as if it were stored and installed on the local machine. To eliminate excessive lag at the application launch, a “streaming” process is employed to push essential components of the application first to permit it to load and continues to download the remaining components of the application as a background process. Once loaded for the first time, the application is cached to the local machine. Successive launches of the application are then loaded from the local copy unless the application has been upgraded. If the application is upgraded, the “streaming” process reoccurs to update the local cache. When the application is running, the local machine's memory resources are used. Since (as noted earlier) the virtualized application includes all shared resources such as DLLs within its own container, the application will use those that are packaged within its container as opposed to the local operating system's installed versions which also may be running on the local machine. Hence, there can be an additional memory resource overhead, but this method essentially eliminates the conflicts between applications as well as provides centralized control of virtualized applications for a more efficient management. Application virtualization can permit time to live stamping for applications which, once expired, removes the virtualized application from the user’s computer.

 

Virtualization Trends in Law Firms

Law firms are traditionally 2-3 years behind the technology trends, and with a good reason. The law firm motto is generally – let the technology mature, let it wean out the weak and establish its strong leaders before committing an investment in such technology. Once “bleeding edge” technology, server virtualization has demonstrated its maturity and it's going to be around for the

long haul. 58% of law firms surveyed by ILTA in mid-2007 report that they use server virtualization. A cross-market TechTarget survey conducted earlier this year showed 64% of respondents using this technology.

View survey here

 

Case Study: Goodwin Procter LLP

In April of 2004, the IT Network Services team was grappling with how to meet the ever escalating systems needs of a large law firm (~1300 users at the time): Servers were rapidly multiplying to meet testing and production needs, datacenter space was filling, and new applications needed to be deployed in the time frame that is our unofficial motto: “Sooner, Rather Than Later.” Noting that many of our servers spend the bulk of their hours idling or operating with very low processor/memory utilization, we felt the tool that we needed was server virtualization.

The mantra at the beginning was “development only.” We’d heard from companies (outside of the legal community) running anything and everything as virtual machines from the very beginning, but we were skeptical and initially proceeded very slowly. The first systems we identified for virtualization were:

 

• Application and database development (including testing for our document management system)
• Citrix testing
• Administrative needs (e.g. server patch deployment)

But as the benefits came into focus, our virtual infrastructure grew rapidly. Within a few months we were putting “minor” production systems on VMs, and not long after that, we would run nearly any system as a VM. As of December, 2007 we have Exchange servers, document management, financial reporting, and other mission critical systems running on Virtual Machines.

 

The Benefits

Green Technology	The impact of consolidation means less cost on hardware, especially with higher ratios of virtual machines to physical server. The true “green” impact will be seen in less power consumption, less emissions and a slowing of the run-off consequences of the world’s desires and needs to keep current with the rapid pace of technology. According to Gartner Research, enterprise technology accounts for 2% of the world’s carbon dioxide emissions. With the lifecycle of the common desktop computer and server being three years for most corporations and firms, think of the impact on the world’s landfills. Over 160 million desktop computers were disposed in 2007.
Management	Consolidation is the most talked about benefit of server virtualization from a cost and management perspective. Fewer servers equates to fewer number of boxes to manage, and less hardware to monitor, maintain and upgrade. The fact that a VM is literally a set of files sitting on a server unearths many less heralded, but no less important, benefits such as server-to-server communication on the same host is at the speed of the motherboard portability. With server-to-server host internal communication, VMs on the same physical host do not to talk to each other over the physical IP network – they communicate directly via the virtual switch embedded in the host. When designing a virtual infrastructure, consider teaming servers on the same host that frequently interact with each other such as contact relationship management and the e-mail system, document management stores and indexers, financial databases and reporting. With portability, the disaster recovery environment does not need to have identical hardware to the production environment as moving a server from one site to another becomes little more than a file copy operation. Thus, the “bare metal backup” operation becomes simpler.
Rapid Deployment – No Screwdriver Required	In a purely physical environment, deploying a new server includes specifying and ordering the hardware, waiting for it to arrive and racking the equipment before any installation of the operating system and software. In a virtual environment, a network analyst can create a VM by assigning the appropriate resources (processor, memory, disk) from the available virtual server pool and setting up the operating system in a couple of hours.

 

The most notable benefit is cost savings. When you consolidate through virtualization, 10-to-1 and 15-to-1 ratios of VMs to physical hardware are common.

 

	Physical Servers Environment	Physical Servers Environment	Virtual Server Infrastructure
Hardware	(1) Quad Core Processor with 2GB RAM, dual-port NIC and (2) 146GB SAS Drives	(1) Quad Core Processor with 2GB RAM, dual-port NIC and (2) 146GB SAS Drives	(4) Quad Core processor server with 32GB RAM, (2) dual-port NICs and (10) 146GB SAS Drives
Number of Physical Servers	10	20	1
Cost Per Unit	$3,867	$3,867	$26,301
Cost	$38,670	$77,340	$26,301
Number of Servers/ virtual machines	10	20	10-20 depending upon server needs

 

*Figures based on pricing of Intel Xeon based servers from a major manufacturer's web site on December 29, 2007. The above does not include costs of virtualization software, which would add from no cost to a several thousand dollars depending upon product, add-ons such as high availability and DR, and support options.

You may be able to realize further savings with operating system licenses. For example,

Microsoft allows the application of a “Windows Server Datacenter Edition” license to a host running virtualization software (even if it isn't Microsoft's) which allows you to run an unlimited number of Windows servers on that machine. Contact your Microsoft reseller for details.

 

The Cautions

There are many cautions that may be more applicable to some than others. In general, poor or lack of technical training, planning, research, design, testing and documentation collectively can be costly to performance, operation, expansion and future staffing transitions. The IT staff must be well versed in virtualization upfront, especially if it is intended to be used as the base layer for mission critical servers and applications. Training of the IT staff who will support this environment should be beyond a “how to” white paper for the product chosen to provide the virtualization layer. Seek out certified training that covers the concepts and the analytics of design as well as the product specific features. Again, this should be accomplished upfront and not in concert with the implementation of the virtual infrastructure.

The statistical analysis of the current operation is part of the fundamental research that will be the basis for decision processes in the aspects of planning and design. Understand that not all servers and applications will be suitable for virtualization such as a fax server or some latency sensitive applications. Key servers such as one of the domain controllers or a DHCP server should remain physical to cover a contingency scenario which involves a problem or corruption at the abstraction layer of the virtual infrastructure.

Proprietary applications should be investigated with the manufacturer as to any known problems as well as with any product support concerns in a virtualized environment. Even if the applications and server profile is fully supported, it is highly recommended to test before taking a physical to virtual (“P2V”) server live.

The importance of good, thorough documentation and information logging from the base configuration through the servers and installed applications cannot be understated as this new layer is introduced to the infrastructure. The documentation of this environment will serve as a foundation for making modifications to backup, disaster recovery and business continuity strategies and policies. Also, the ease of provisioning servers without the physical limitations invites server sprawl in the virtual environment. By keeping careful and meticulous notes of the environment's evolution, more informed decision processes can take place on licensing, discovery, future growth and change to the infrastructure environment.

These six principles (technical training, planning, research, design, testing and documentation) must be proactively addressed with measurable qualifiers that can be verified before any P2V transition occurs with either a third party integrator or through in-house IT. The development of good implementation habits, strong discipline and business foresight will awaken awareness of the road ahead when the new concerns introduced by virtualization begin to infiltrate the infrastructure.

Learn from others who have “been there, done that.” Their insight on the encountered pitfalls,workarounds and lessons learned are often not documented and can save hours of research and frustration.

 

Hardware Requirements

Virtualization for servers can be deployed on practically any hardware manufactured from late 1999 technology (Pentium III Xeon) to present. However, we don't advocate running virtualization on “vintage” equipment. The profile of the hardware will limit how much and how effectively you can run multiple servers. The traditional approach to server hardware sizing in the physical server world was to include provisional overhead to accommodate peak hours of operation. However, the more prevalent time-span is the non-peak hours where average resource utilization was generally 15% or lower. In the consolidation of the virtualized world, the objective is to get more performance out of the server hardware and keep utilization at or slightly above 60%. The provisioning of additional hardware resources to a throttled server can be done on-the-fly. In a computer network, the environment will only be as responsive as its weakest component. There is no one simple rule-of-thumb formula for how much hardware is needed for "X. number of servers. When moving from P2V, the hardware specifications should be based upon what servers are targeted for virtualization and the analysis of memory, CPU and input/output usage as well as disk storage needs for the targeted severs. More is usually better for memory, CPU cores and I/O components, but you don't need to “buy a Cadillac if a Buick will do.”

Consider the other factors such as High Availability (“HA”) as part of the plan for business continuity to minimize downtime in the determination of what hardware is best suited for the current environment and anticipated growth. Hardware expandability is key as well as purchasing or having like servers, especially if HA is part of the plan (the same motherboard and CPU chipset will ensure full compatibility in a pooled resource environment).

It is usually best to buy new servers to be the underlying fabric for the virtualization infrastructure. It is always easier to justify the migration to virtualized server when a significant number of the existing hardware for servers are slated for retirement.

Desktop virtualization will also work on hardware manufactured from late 1999 technology (Pentium II) to present with at least 512 MB of memory. Virtual desktop machines running under a desktop perform better with faster CPUs and more memory (most vendors recommend 2 GB of system memory for their desktop computer VM products).

 

Server Virtualization – Concept to Implementation

The following three sections provide a high-level overview of the process to migrate physical servers to a virtual infrastructure.

Analyzing the Infrastructure.

The most important part of migrating from a P2V environment is the planning and design phases. First, an in depth detail of the existing environment as well as the analytics of the current operation must be discovered and documented. This will help answer the question as to the hardware profile needed to efficiently run the virtualized servers.

Prepare or audit documentation of existing servers. Each server's documentation should consist of a profile to include at a minimum:

 

CPU	number of processors, model, onboard cache and clock speed
Disk Storage	quantity in use, number of disks in each volume, types of drives, growth rate and RAID configuration
Network Adaptors	type of NIC, bandwidth
Memory	quantity and type
Operating System	build and version
Services	What is installed as which services are running
Applications	All installed applications – build and version of each

 

Run monitoring on the targeted P2V servers to gather the current operating statistics during normal and peak periods that should include the following:

CPU	Processor\ % Processor Time (all instances)   System\ Processor Queue Length (all instances)
Physical Disk	Disk Transfers/sec (total IOPS for the volume)   Disk Bytes/sec (total data throughput for the volume)   Avg. Disk Queue Length (for each windows disk this should be less than 2 plus the number of disks, 3 Disk RAID 5 should be less than 5)   Avg. Disk Sec/Read Read I/O Latency of the volume Measured in Seconds. (typical values are in milliseconds)   Avg. Disk Sec/Transfer I/O (Read & Write) Latency of the volume Measured in Seconds. (Typical values are in milliseconds)   Avg. Disk Sec/Write Write I/O Latency of the volume Measured in Seconds. (Typical values are in milliseconds)
Memory	Memory\ Available Mbytes   Memory\ Pages/Sec   Memory\ Cache Bytes
Network	Server\ Bytes Total/Sec   Server\ Work Item Shortages   Server\ Server Sessions   Server\ Logon/sec

 

Trends in server use and disk storage needs should be considered to ensure there is ample overhead factored in for future growth in the virtual infrastructure.

 

Planning and Design

As pointed out in the section above, the process begins with the gathering of statistical information and the assessment of the existing environment. Think through the entire implementation process capturing the objectives, thoughts and understanding of how the virtualization change will impact the infrastructure and the management thereof. Prepare a detailed project plan before engaging in any work or purchasing. The more detailed the plan, the lower margin for error or miscalculation on the implementation, costs and subsequent management needs of the new environment. It is highly recommended to run a test environment of each server before going live. A test environment can exist on the same hardware infrastructure as the production environment.

The project plan and infrastructure design should address the following questions:

• Is there a need for new hardware?
• What class of hardware will make up the building blocks of our virtual environment – Blades? Chassis servers?
• What processor platform will provide the best performance?
• What servers will be virtualized and where will they impact operations while being transitioned, and are there any dependencies?
• Will the virtualized environment have DR and HA for the virtualized infrastructure and to what degree?
• What criteria will be used to validate the success or failure of a virtualized server?
• How will backups be performed and take advantage of the virtualized environment?
• How much automation will be added for DR and HA?
• What modifications need to be made to the DR and Backup policies?
• How will the environment be managed and monitored?
• How will virtual server provisioning be controlled and monitored?
• How will upgrades and replacement in hardware, firmware and utility software be applied to the virtualization platform?
• What tools will be used, installed and deployed to proactively monitor the operation?

The plan should be organized into a logical, chronological schedule of the above points to include checks and verification steps. The underlying virtual server framework is the first piece to be built.

The overall plan should also include the impact concerns of the system users throughout the process – how will users be impacted? Proactive communication to all areas of IT throughout the entire process is key to lessen frustration and help overcome the resistance to change.

 

Implementation

The best approach for the migration to a virtual environment is through a gradual and methodical implementation. The project plan should be used as the guideline for the implementation to ensure all aspects of the migration stay in check. The recommended steps of implementation are as follows:

 

Build	The baseline installation or P2V migration. Document each step along the way as a diary for the migration. This may need to be re­examined or possibly recreated during the test and deployment phases of implementation.
Test	Run the virtual server in a test mode and validate that all criteria, features and fault tolerance fall within expected guidelines for operation. Test all DR, HA and backup/restore aspects early on and before any deployment to ensure full functionality of these important areas of the operation.
Deploy	Move test environment (or in some cases redo the P2V migration) to the production environment.
Manage	In the virtualized environment, tighter controls are necessary to achieve a higher volume of consolidation. Thus, there is a need to find better ways to proactively manage and monitor the servers. Make sure there are means included to equally measure and proactively monitor performance post migration. Thresholds should be closely monitored and statistics logged to justify tweaks made to settings and configurations of the servers and underlying virtual framework.   Patch and upgrades to operating system and application are also in included in the management process. Ensure defined policies and procedures are updated to reflect and take advantage of operational changes with the enhancement of the virtualized environment.

 

Conclusion

Today, the maturity of virtualization has far more benefits that outweigh the downsides. Virtualization is fast becoming the standard for datacenter infrastructure, application deployment and desktop deployment. It should not be overlooked in any strategic planning of the IT operation.