Like mops from Disney’s “Fantasia,” the number of servers deployed since 1997 has exploded, with many companies doubling or tripling the size of their production and non-production server farms. The factors driving the increase include massive e-commerce initiatives, declining server prices and business process re-engineering projects.
This spawning of servers has had a significant and somewhat hidden downside: management cost. The servers themselves are expensive and the cost to operate them typically represents 60%-70% of their total cost of ownership. This would not be a problem if the servers were highly utilized and providing significant value, but often that is not the case. In fact, there is empirical and anecdotal evidence that points to the opposite:
* Servers typically operate at 10%-15% utilization.
* Close to 50% of the servers purchased are Wintel boxes with one or two CPUs.
* Many servers are single-purpose.
When one considers these facts in conjunction with the pressure on CIOs to do more with less, a simple question arises: Can one reduce capital and operational expenses by consolidating servers onto larger, highly utilized boxes without sacrificing performance or reliability?
The answer is yes, and many CIOs are doing just that. Findings from a comprehensive study that analyzed the server consolidation trend earlier this year by the Gartner Group concluded that 69% of interviewed companies are undergoing server consolidation, 25% are considering it and 6% have no plans to consolidate.
Historians may ask, “What’s the big deal? IBM S/360 mainframes have enabled workload consolidation and partitioning for 30 years. Isn’t this back to the future?” In a way it is. What is new this time around however is the architecture. For the first time, large-scale server consolidation is now being considered for distributed Unix and Windows environments.
In this viewpoint, we will investigate the various ways private and public enterprises are consolidating their distributed server farms, the benefits associated with that consolidation and some of the technologies enabling this trend.
Firstly we must define exactly what server consolidation is. John Phelps of the Gartner Group classifies server consolidation into the three categories: logical consolidation, physical consolidation and rational consolidation.
* Logical consolidation is the process of identifying all servers and server resources within the enterprise. Though this may sound simple, most enterprises do not have a firm grasp over how many servers they own, where they are located, what role they play, how frequently they are used and who manages them. This logical consolidation of knowledge allows the IT organization to better manage the server infrastructure, enabling more intelligent acquisition and usage strategies.
* Physical consolidation is reducing the number of physical sites where servers are installed and managed. This is typically achieved by consolidating data center locations internally or co-locating servers at a third-party data center. By reducing the number of sites, an IT department can improve its control and service of service and reduce real estate expenses.
* Rational consolidation is the most complex and entails taking a single server and dividing it up into multiple logical servers. Rational consolidation is further divided into two sub-categories: workload consolidation and partitioning
* Workload consolidation assumes that there is one instance of the operating system with many applications and workloads sharing the underlying physical resources managed by that one OS. In effect, workload consolidation technologies improve the OS scheduler, a critical task as most schedulers (particularly Windows) are designed so that one physical server supports one OS and one application. Once you place multiple applications on the same box, each with very different workload requirements, a traditional scheduler can’t keep up. Consolidation technologies can reduce capital and operational expenses, but they suffer from one major drawback: isolation. If one application fails and brings down the operating system, all of the other applications will crash as well.
* Partitioning assumes that there are many instances of the operating system, each supporting its own application and all of them running on a single host operating system. In other words, a server would run a single version of Windows and the partitioning technology would allow a systems administrator to run multiple virtual servers on top of that host OS, with each virtual server capable of supporting it’s own instance of an OS and an application. Partitioning reduces capital and operational costs just like workload consolidation-with the added benefit that each virtual server is isolated from one another. If one application and OS crashes, it will have no impact on the other applications running on that server.
The ROI of Consolidation
Because server consolidation for distributed Unix and Windows environments is a relatively recent trend, there is little empirical data of hard ROI savings. Battery Ventures, however, has discovered significant anecdotal ROI data by conducting research in the server consolidation market. Some of the highlights of this data include the following:
* A Fortune 500 company has 300 servers in its IT organization handling development, test and QA functions. It says it expects rational consolidation of its servers to cut this number in half, saving between $5 million to $10 million over the next five years.
* A Fortune 500 company has 175 single-CPU Wintel servers handling print functions. Through rational consolidation, it says it anticipates that it will cut that number by close to 70%, saving more than $5 million in capital and operating expenses over the next few years.
* A Fortune 50 company says that one of its divisions can eliminate 600 development and testing servers from its server farm, saving $2.4 million per year.
Beyond the savings on capital and operational expenditures, reducing the number of servers to a manageable number saves corporations money by cutting the amount they spend on real estate, power consumption, security, and administration.
Physical and logical consolidation initiatives are 10% technology and 90% process. Though a number of systems management, asset-tracking and discovery tools can help IT organizations better consolidate their server resources, the success of these initiatives primarily depends on IT organization vigilance to enforce procurement, deployment and change management processes once the servers are consolidated. Without agreed-upon and enforced policies, the best asset-tracking tools will do little to keep consolidated server farms under control. That said, leading vendors of discovery and asset-tracking technologies include BMC, Computer Associates, Hewlett Packard, IBM/Tivoli and NetIQ.
Some of the most exciting technology supporting server consolidation is happening at the rational consolidation level and, in particular, with partitioning. In fact, partitioning can be accomplished either through hardware or software.
Hardware-based partitioning requires the server vendor to pre-define the configuration in which resources can be allocated. For example, in the Sun E10000, Sun has defined the lowest common denominator for partitioning to be a Quad Board with four CPUs, four I/O paths and 4GB of memory.
Though this allocation is useful in defining pre-set and isolated configuration limits, it is inflexible since systems administrators need to operate within the defined configuration parameters and resources cannot be shared across partitions. Hardware-based partitioning technologies are being built by large server vendors like IBM, Sun and Unisys.
Software-based partitioning requires the enterprise to run virtualization/emulation software on top of the host operating system and intermediate the communication between the host OS and virtual servers. The key benefits behind a software-based approach are that systems administrators have more control over physical resources and can share those resources across each partitioned environment. Fractional processors, memory, and I/O paths can be shared across many applications.
There are a handful of companies emerging to create this software partitioning technology-although the market is still young and most customers are deploying it cautiously. In fact, most customers we spoke with say that they will first test partitioning technology in a non-production environment. Some of the leading vendors of software based partitioning technology include private companies like Aurema (backed by Intel Capital), VMWare (backed by Azure Capital, Goldman Sachs, Dell Ventures and JPMorgan Partners), SWSoft and Connectix. Battery Ventures has not yet made an investment in this space, but is actively evaluating several such opportunities.
Server consolidation for distributed Unix and Windows environments appears to be gaining momentum, due to the very compelling potential to cut capital and operational expenditures. Most enterprises are dipping their toes in the water, starting first with logical and physical consolidation. Over time they will move toward rational consolidation, where they can achieve the greatest savings.
If the server consolidation trend takes off as expected, the companies providing enabling technologies stand to be large beneficiaries. Entrenched players like Computer Associates and IBM have a leg up, but there may still be room for a startup or two with a better mousetrap.
Roger Lee has been a senior associate with Battery Ventures since 2001. He focuses on investments in enterprise software and Internet infrastructure. He was previously co-founder of managed service provider Corio Inc.