ESXDeployment.doc

© Copyright 2007 GlassHouse Technologies, Inc. All rights reserved.

Accelerate your ESX Deployment

Introduction

You know server virtualization is a good idea. The benefits are significant, both from a capability and financial standpoint. Yet many organizations have not yet moved to take advantage of this technology on a large scale. The purpose of this paper is to both illustrate the benefits of moving quickly to build your VMware ESX-centric infrastructure, and to suggest a framework that will help you to accelerate your deployments.

We have assisted many organizations with the design, planning, and deployment of their ESX infrastructures. In this paper, we share the common themes that have developed throughout these initiatives. It is our hope that the information provided here will save you time and allow you to accelerate your deployment.

Hard-dollar Cost Savings: The Reason to Accelerate

Like all infrastructure improvement initiatives, return on investment has historically been difficult to deliver from an IT infrastructure project. However, when using VMware ESX Server to contain server proliferation and consolidate current underutilized servers, the numbers tell a very positive story.

The following case study from a large financial services company shows the hard-dollar savings that can be realized by virtualizing using VMware ESX Server. The company anticipates many benefits beyond those captured in the ROI analysis that follows. However, the difficulty in quantifying those benefits and having them stand up to the scrutiny of the organization’s financial analysts led the company to focus initially on the easy-to-quantify savings illustrated in our analysis.

CASE STUDY

This particular financial services company relies heavily on Windows and Linux servers within its three nationally distributed datacenters. The company’s IT infrastructure team has determined that as many as 80% of the organization’s servers are potential candidates for VMware, based on the data sampled in the environment. The company engaged us to help

1

GlassHouse Whitepaper

The cost benefits of server virtualization demand your attention. Server virtualization is the most significant opportunity available for cost savings in the datacenter. It is not a question of if you will use the technology, but rather how quickly your organization will begin to realize the benefits.


develop its virtualization strategy and, subsequently, its design and deployment plan.

2


OVERVIEW

Currently, the company has 1,112 Intel servers spread between its datacenters. Of these, approximately 80% are targeted to move to virtual machines. The company refreshes its servers approximately every three years. Fixed asset discovery found that more than 35% of the target physical servers are scheduled for refresh this year, and approximately 40% are scheduled for refresh over the following two years. This refresh includes upgrading to Windows 2003 during the migration process.

There are also a significant number of servers on relatively new hardware that do not require a refresh in this time period. To accelerate the benefits of virtualization, the team has targeted them for the virtual environment. After moving these servers to VMs, this newer hardware will be repurposed to be used for any new physical servers that may be required.

THE DESIGN

The design calls for separate ESX farms for every nine quad-processor ESX hosts. Eight hosts are expected to be fully loaded. One host is set aside as a “swing” server. The swing server acts as a host to house the initial deployment of VMs that require base lining, and also acts as a recovery server should one of the other eight servers in the farm fail.

The team expects to achieve a production ratio of approximately 20-24 VMs per fully loaded host, or between 160 and 184 VMs per farm. Because the swing server hosts only a few VMs, the overall host to VM ratio will be approximately 20:1.

From a SAN perspective, the design team defined several classes of storage divided into tiers. Tier 1 storage is intended for critical applications that require back-end replication between datacenters. Tier 2 storage is based on a high-end frame that allows local snapshots but does not include the cost of replication. Tier 3 provides local SAN storage only, with no snapshot or replication technology.

Most VMs are expected to use Tier 2 storage. Each ESX host is connected to the SAN via two separate Fibre Channel adapters. There are four total Ethernet network connections.The farms will be managed primarily using VMware’s VirtualCenter. Backup and monitoring agents will be installed into virtual machines to support applications that require them.

FINANCIAL ANALYSIS

3


From a financial point of view, this project is very similar to the majority of production virtualization projects. Application initiatives and growth drive the need for additional servers in the environment. At the same time, the organization has many underutilized Intel-based servers that would be extremely difficult to consolidate within the native OS. This makes virtualization an extremely attractive strategy to lower the overall cost of the infrastructure.By looking at just the hard numbers associated with one-time procurement

costs and the ongoing hardware power, cooling, and space costs, it is obvious that virtualization of new servers, servers ready for refresh, and P2V candidate servers not yet up for refresh all create very favorable return on investment based on any standard metric, including IRR, NPV, or payback period.

ASSUMPTIONS FOR THE MODEL

Figure 1: Assumptions

Figure 1 defines the settings used for the financial model.

Project Cost Avoidance

Based on the assumptions, Figure 2 illustrates a comparison between environments with all physical servers versus an environment where 80% of the servers are virtualized.

The organization would save approximately $6 million if the environment were to be built new today. Overall, the project would require 222 physical servers, plus 45 ESX hosts. The ESX hosts would support 890 virtual machines over five farms, requiring approximately 18 TB of SAN space.

4


ONGOING COST SAVINGS

Figure 3 illustrates the differences in power, cooling and hardware support on a yearly basis.

PHYSICAL VS. VM COSTS

Figure 4 illustrates the cost differences between physical and virtual servers, based on one-time and ongoing costs for the organization.

From a high level, an important cost to note for the organization is the $11K required to deploy a physical server versus approximately $4K to deploy a virtual server.

Note also that providing power, cooling and basic hardware support for the current physical servers costs more than $2K per server annually, versus $129 annually for a comparable VM.

Figure 2: Project Cost Avoidance

Figure 3: Ongoing Cost Savings

5


Figure 4: Physical vs. VM Costs

CASH FLOW ANALYSIS

The following analysis compares the continuous use of physical machines for new deployments and refreshes versus using VMs to provision 80% of the new servers, server refreshes, and servers to be P2V’d. The design is defined in The Design section on page 5. Year 0 indicates the initial project outlay. The analysis covers five years, but both three- and five-year metrics are shown.

In Figure 5, the virtual environment includes 326 VMs in Year 1, followed by 204 in Year 2, 220 in Year 3, 70 in Year 4, and 70 in Year 5. These numbers are split between new servers, servers being refreshed, and servers being P2V’d. The cost to provide this capacity is derived from the costs defined in the Project Cost Avoidance section on page 6.

The project-related costs are shown next. Here the investment is just over $300K. This includes $65K in consulting to develop the virtualization technical design and deployment plan. It also includes the labor required to build the infrastructure and to develop processes, procedures and other documentation. The other project-related cost is for the labor to P2V the servers that will not have upgraded operating systems.

The last part of the virtualization section in Figure 5 is Hardware Support. The cost for this is placed into the following year, assuming the migrations have been performed throughout the year. Hardware support is very favorable for virtualization because there are significantly fewer servers in the virtualized model.

The physical environment is relatively simple. The cost to provide servers equals the total number of servers required per year times the cost to

6


implement a physical server. Hardware support costs are calculated in the same fashion for both the physical and the virtual environments.

The hardware costs and support costs are then added together, and the two models are subtracted from each other to create the model’s cash flow. This cash flow is the amount that would have been spent if virtualization were not implemented.

Figure 5: Cash Flow Analysis

ROI METRICS

We used The Present Value of

Cash from the cash flow analysis in Figure 5 to calculate ROI metrics. Figure 6 breaks out these metrics. As you can see, this project is extremely favorable using any of the ROI metrics.

Figure 6: ROI Metrics

CONCLUSIONS FROM THE CASE STUDY

7


Rarely is a new technology as easy to justify financially as a virtualization project using VMware ESX Server and VirtualCenter. The numbers on the previous pages speak for themselves. And remember, this is a purely cost-based analysis. We didn’t even consider the potential cost savings your company would see in availability and disaster recovery.

Given the magnitude of saving gained by virtualization, it would be imprudent to wait to implement this technology. It would also be imprudent to deploy only when servers require a refresh. It is a far superior approach to standardize the infrastructure on virtualization and plan to migrate all possible candidates to ESX as soon as possible. The more accelerated the deployment, the quicker and more substantial the payback.

A Framework to Accelerate Your Deployment Successfully

CANDIDATE DISCOVERY

It is imperative that you perform a complete candidate discovery process to identify opportunities for the consolidation of your organization’s existing servers. The greater the number of candidates you identify, the greater the ongoing cost savings toyour organization.

The candidate discovery process should include the following: Creation of a server matrix containing every Intel server in the

environment, including:

o Server Hardware Inventoryo Server OSo Applications or running serviceso Server fixed asset information (depreciation)o Performance information on core resources for each server on

the listo Disk in use and allocated for each servero Network or site location informationo Tier level of the server for recovery SLAso Any special considerations that would rule out a candidate

(special hardware, etc.)

Performance data for the core resources should be sampled over three to four weeks, taking into account business and peak hours. The AOG CapacityPlanner™ tool is the most effective way we have found to capture this information. Keeping in mind your business

8


objectives, experienced VMware experts will correlate and analyze this captured information.

Companies often perform partial discoveries or “assessments” by discovering and analyzing a subset of their server environment and then extrapolating this information to determine the potential for server virtualization. This extrapolated sample helps them build justification for the project, and can be a valuable tool when used in this way. However, do not make the mistake of confusing an incomplete discovery with an actual detailed candidate discovery process.

DESIGN

Often, organizations’ initial forays into server virtualization have been for spot solutions rather than a design with the “big picture” in mind. Using ESX as the base for your strategic infrastructure will certainly require a big picture view. Understanding design alternatives and the benefits and drawbacks of the alternatives is paramount when building an infrastructure that will meet your company’s objectives.

Examples of some key design alternatives include: Storage:

o How will we connect to the SAN?o LUN sizing / storage requirement determinationo Tiered storage for VMs?o What will reside on the SAN versus what should reside on

the ESX host?o Effective use of snapshots/mirrors and replication

Hardware selection—should you choose duals, quads, eight-ways, or blades? What is the cost-effective “sweet spot” for your organization?

Network connectivity:o Number of NICso Type of load balancingo Speedo Bondingo Virtual switch designo Network security zoneso Backup networkso Leveraging backup networks for VMotion

Memory configuration

Processor architecture selection—dual core versus single core

Tiered farm design

Workload design and selection

Security

9


Recovery models for both ESX servers and VMs

Resource allocation standards

Impact on IT deployment and management processes

Monitoring

Capacity management

VirtualCenter integration into existing models.

As you can see from the preceding list, an ESX implementation will impact every facet of the IT infrastructure. Therefore, it is important that all affected groups have a voice in the design. It is important to gain consensus on the overall strategy to keep the project from being sidelined. Representatives from the following groups should participate in the development of the design:

Security Networking Storage/backup Intel engineering Application/ server owner representation.

DEPLOYMENT PLANNING

Once you’ve designed the ESX infrastructure, you must then plan the deployment of the technology. Without a plan, technology is often pieced together as time and resources allow. Given that this technology touches all aspects of the organization’s computing infrastructure, you should make a concerted effort to ensure that your organization follows proper implementation steps.

You will find the following stages in almost every VMware deployment (in most cases, several of these stages can run concurrently to allow you to begin deployment sooner and realize the benefits of the new technology faster):

PROCUREMENT

Generally, the procurement of hardware and software for the ESX infrastructure is on the critical path of the project. A comprehensive bill of materials that details all products required for deployment should be developed at the conclusion of your design session and should be used to facilitate rapid acquisition of the hardware and software. Keep in mind that any internal procurement processes that have been put in place to ensure appropriate checks and balances will slow you down if you are not familiar with the processes and prepared with all the necessary authorizations. Even

10


if you’ve done a good job demonstrating ROI and selling the benefits, you will likely be asked to prove savings on a continual basis.

LAB/TESTING

This stage of the project contains tasks that allow the team to test out the design components required to build the base infrastructure. Engineers will develop and test base configurations, backup configurations, scripted installs, VM provisioning, etc. Any testing required for basic technology reasons—or any required POCs—should be executed during this stage.

BASE INFRASTRUCTURE BUILD

During this phase, the first production ESX servers and the supporting infrastructure required for your initial VM capacity is put into place.Components of the base infrastructure build include:

Deploying and configuring ESX hosts Deploying and connecting to the storage solution Establishing network connectivity Implementing monitoring and management Deploying VM templates

This phase of the project can often move fairly quickly if the processes and steps for deploying the solution have all been developed and tested in the Lab / Testing stage. It is important to note that the base infrastructure in place at the end of this stage it is not yet “production-ready” until integration, monitoring, and management processes have been developed and instituted.

PRODUCTION-READY PROCESS DEVELOPMENT

In the interest of accelerating your deployment, this phase should run in parallel to the Lab/Testing and Base Infrastructure phases. Here, the deployment team will be developing or modifying existing processes to make ESX and virtual machines fit into the current production environment. Once the processes are in place, the infrastructure can be deemed production-ready for VMs.Examples of processes that will need to be developed or modified include:

Cost model for initial provisioning of VMs Chargeback model modifications Physical server P2V processes VM decommissioning processes V2P process for support/risk mitigation Change control procedure modifications

11


Monitoring, alerting and notification processes Troubleshooting—creation of support scripts Responsibility charts Modification of server requests and deployment processes to take

VMsinto account

Access methods possibly modified for application owners and engineering

Resource allocation and standard VM configurations standardized and made public

Upgrade and patching processes.

PRODUCTION-READY

Figure 7: Possible Deployment Process

Once the base infrastructure is in place and the appropriate processes have been instituted, you are ready to deploy and manage new VMs in your ESX infrastructure. Going forward, all new servers should be deployed via the process diagrammed in Figure 7. The idea here is to presume that any new server will be a VM candidate unless there is a reason it cannot be provisioned as such. This is critical to server containment.

P2V MIGRATIONS

Likewise, migrating existing physical servers to VMs is also key to server consolidation and the associated ongoing cost savings. The three most important elements in the P2V migration are:

Proper discovery to ensure that the migration does not get stalled Proper allocation of resources to ensure target migrations are met

12


Proper project management and scheduling to ensure servers and application owners are available and ready when scheduled.

Keep in mind that P2V migrations and subsequent cost savings are often the key component of a cost justification. As such, every month a server is not migrated costs the company money.

The P2V process will include:

Completion of server discovery and candidate selection Coordination (scheduling/notification) of timing Preparation of the source server Conversion of the physical machine to virtual disk files. (You may

benefit from the use of an automated P2V conversion tool for this step. Bear in mind, however, that although the conversion itself can be automated, the due diligence associated with the P2V still requires significant effort.)

Functionality testing/QA Go/no-go decision Retirement of the physical server (its unplugging and removal to

realize power, cooling, and space savings).

During discovery and planning for a large number of P2V migrations, it is important to recognize the fact that there are often different risk levels associated with P2V candidates. By assigning a risk level to each candidate, you can perform the appropriate level of due diligence.

A good example of a low-risk candidate server would be a print or application server in an environment that has already migrated similar servers successfully. In this case, the candidate server may be migrated and QA’d for functionality and released to production, after which the physical server can be retired. Total estimated time is about three to four hours per candidate.

Conversely, an example of a high-risk server would be a candidate that hosts a unique application that has a high level of visibility in the organization and is potentially sensitive to hardware changes. In this case, you should perform exhaustive testing prior to retiring the physical server. See Figure 8 for a graphical representation of a high risk versus low risk P2V deployment.

13


Figure 8: High Risk vs. Low Risk P2V Deployment

Key Considerations

THINK “BIG PICTURE”…DON’T BUILD A BUNCH OF SEPARATE SILOS

ESX is often implemented in small spot solutions prior to being implemented on a large scale. If not held in check, this deployment method can lead to numerous hard-to-manage, difficult-to-troubleshoot, and disparate systems. When designing your environment, think ‘Big Picture.’ Although groups of servers can be deployed at different stages for different reasons, having standardized processes and configurations will let you manage, monitor, and deploy your environment as a single cohesive solution.

BUILD BASE INFRASTRUCTURE CAPACITY FOR YOUR PLANNED REQUIREMENTS AND DESIGN IT TO SCALE

14


Often, the driving force behind an initial ESX implementation is to meet a date to enable another project to use VMs. This is a good way to seed the environment and to start the infrastructure build. The key to success here is to design the solution to expand and support more than just the initial project. A properly designed ESX environment should be able to scale and support numerous projects, and hundreds, if not thousands, of VMs.

P2V AS MANY AS MAKES SENSE

Organizations often use ESX to stop server proliferation. However, don’t hesitate to look for other virtualization opportunities. In most environments, numerous servers exist that could be virtualized with a positive cost benefit to the company. Some of these opportunities include:

Under-utilized servers on newer hardware. Virtualize these and return physical assets to the server pool for use in a project requiring a physical server.

Older servers due for a refresh. P2V these to show a cost avoidance on new server purchases.

Servers near the end of their depreciation schedule. P2V these prior to lifecycle when the annual savings are larger than the remaining depreciation.

Datacenter space constraints. P2V as many servers as possible to avoid datacenter expansion costs.

In short, a company should P2V as many servers as fiscally feasible. If removing a physical server from the environment saves the organization money, the migration should occur.

STORAGE IS A KEY COST COMPONENT OF THE DESIGN

Storage costs for an ESX implementation are almost always the single largest line item in a VMware ESX bill of materials. Establishing your requirements early and designing a storage solution for ESX that meets your SLAs will be one of the most important decisions you make.

IMPACT ON CHARGEBACK MODELS

Organizations that charge back users or business units by the processor or by the machine will need to make changes to their chargeback systems. For whatever reason, this tends to be looked at as an insurmountable task. The answer to this is simply to use a fraction multiplier of the organization’s current model, or to offer a different price for the VM. The biggest draw to VM technology is the cost savings. It makes sense that if the business is charged the same dollar amount for a VM or a physical server, it will opt for the physical server.

PRODUCTION AND DEVELOPMENT: TOGETHER OR SEPARATE?

15


In a traditional server environment, the combination of production and development servers on the same network sharing resources is considered unacceptable, and for good reason. However, VM isolation eliminates the reasons these environments have traditionally been split. Heterogeneous workloads have a number of benefits in a virtualized environment. Some of those benefits can include:

Higher VM:Processor consolidation ratios Better time to market (ability to move servers from development to

production almost instantly) Test/Development VMs can be shut down during a recovery scenario,

leaving capacity that can be used for production VMs without reconfiguration of the host servers

Many of our clients are mixing production and development VMs on the same ESX hosts to take advantage of these benefits.

DESIGN CONSTRAINED BY CURRENT THINKING AND EXISTING STANDARDS

Existing standards for servers, network components, SAN connectivity, etc. often limit design teams from developing the best solution. Standards that have been in place for physical Windows/Intel servers might need to be revisited to ensure that they are not impeding an effective deployment.

16


Conclusion

The cost benefits of moving quickly to a VMware ESX-centric infrastructure that we have illustrated in this paper are typical across the various organizations and industries that we’ve worked with. Also typical is the 80% number when we look at opportunities to virtualize. We encourage you to look at your server environment and apply the metrics in this paper to get a better idea of the opportunities in your organization.

Because the benefits are so significant, a diligent, big-picture view is essential to actually realizing the benefits of designing, planning and deploying your ESX environment.

17


About GlassHouse Technologies, Inc.

GlassHouse Technologies is an independent consulting and services firm focused on transforming IT infrastructure to a service provider model. TransomSM, GlassHouse’s proprietary methodology, aligns business processes and information technology systems, transforming our client’s existing infrastructure into scalable, compliant, cost-efficient and tightly organized environments. Whether focused on data centers, infrastructure optimization, data management or managed services, GlassHouse consultants architect, implement and operate IT environments to drive high performance and agility. GlassHouse clients include Morgan Stanley, State Street Global Advisors, Allianz, AIG, Tel Aviv University, Reed Elsevier, Zamil Steel, Amgen, Biogen, Turk Telekom, Virgin Atlantic, and Israeli Defense Forces.

18

ESXDeployment.doc

Technology

organizations servers

server virtualization

usingvmware esx server

benefits of virtualization

currentunderutilized

intel servers

server proliferation

recovery server