Goldilocks the IT guru – determining the correct level of IT resources

Take a look at how just-in-time manufacturing and distribution processes have revolutionized supplier operations, eliminating the need to keep too large a store of raw materials and finished goods on hand. But this trend can be taken too far – not having the required resources when and where needed halts production and drives away customers. To resolve this, companies have deployed extensive economic forecasting and supply chain suites to ensure they can get what they need when and where they need it.

The same applies to computer systems. By knowing exactly how much computing resources it will need at any given time, and how current resources are being utilized, a company can reduce its capital budget and cut operating expenses without reducing service levels. As with supply chain optimization, it requires accurate capacity planning to ensure resources are sized just right.

Too Big or Too Small
Ineffective capacity planning presents two problems. The most obvious one occurs when an area is underprovisioned. In this case, networks slow down, customers can't get access, processes time out, backups fail and help desk calls escalate. The standard response to the problems of underprovisioning is to overprovision – buy more resources than one needs. In some cases, this makes sense. For example, in building out a network, there is minimal price difference between 10/100 Ethernet and Gigabit Ethernet even though there is a tenfold difference in bandwidth. It is better, therefore, to deploy Gigabit Ethernet even though one doesn't need the extra capacity immediately.

In other areas, however, it is a costly, senseless way to work. Yet, companies do just that. The late nineties were famous for overprovisioning and, although many companies have gotten smarter, it is still common to find servers running at twenty percent of capacity. It is essential to have some overhead built into the system to accommodate rising demand and peak traffic loads, but excess capacity does more than waste the capital expenditure budget. Each piece of equipment also has ongoing expenses including:

• Operating system – Each server requires OS license payments, even if one is using a supported version of Linux.
• Application licenses – These are often sold on a per processor basis, so the more machines you are running, the more licenses you pay for.
• Management software – These often require that an agent be installed on each monitored device, with a payment for each agent. The more devices, the more agents, the more payments.
• Power and cooling – Whether you are environmentally conscious or just want to save money, cutting the number of servers makes sense.
• Personnel – It takes more time to maintain a large amount of underutilized equipment than just the right amount.

To combat misprovisioning, several strategies exist to correctly size equipment. One is to take the modular approach – using a large number of small, commodity servers which can be scaled out as needed. When demand increases, just buy some more servers and bring them on line. Then there is the opposite tactic – server consolidation. Here, the goal is to reduce the number of devices. Instead of having separate machines each devoted to running a particular process, and each with its own overhead, processes are brought together to run on a single large server or mainframe with shared resources. While the initial equipment costs are higher, this approach generally results in reduced ongoing expenses.

Some server manufacturers take a hybrid approach, shipping devices with extra processors. The customer only pays for the additional processors if they start to use them. This has been used for a while with mainframes and is now starting to extend down into Intel-based machines. In June 2005, for example, Unisys Corporation released the Real Time Capacity (RTC) series of its ES7000 servers. The ES7000 RTC servers, which run either Linux or Windows, come with 4, 8 or 12 active processors as well as four inactive processors. When they need added capacity, customers can activate one or more of these extra processors using either a 15-day or a permanent license.

Finding a Fit
While developments such as server clusters or Unisys's RTC are steps in the right direction, they are severely limited in scope. In today's connected world, delivering service is not tied to a single piece or type of hardware. Instead, for example, a single service may require pulling information out of a mainframe database, processing it through an application residing on a mid-range Unix server, and then delivering it to the end user via a commodity Linux web server. Further complicating the matter, more than one application will be utilizing that database, running on that Unix server or being served to users over the Internet. Assuming the service problems of underprovisioning are out of the question, this leaves companies two choices: costly overprovisioning or accurately predicting the correct capacity level. This is the role of capacity planning.

Capacity planning, while traditionally associated with hardware purchasing, is actually an ongoing process of assessing current business needs, IT service levels and ensuring the proper level of service is available at the right time to meet future business goals. And, of course, to do this in as cost-effective a manner as possible. Its uses include:

• Consolidating applications onto a single server.
• Determining whether a service should be provided in house or outsourced.
• Planning technology migrations. For example, if the company upgrades its Windows version, it isn't as simple as just loading the new software on the existing equipment.
• Establishing a disaster recovery system. Do you need to set up a fully-redundant backup environment, or can existing equipment fill the need?
• Facilitating business unit plans. If accounting wants to roll out self-service bill payment, do you have the necessary bandwidth and transaction servers?
• Accommodating business acquisitions, moves or restructuring.
• Cutting costs.

There are several ways of accomplishing capacity planning. The simplest is linear trending – taking a look at the level of resources the company has used in the past and using this to project future needs. This is an easy method that doesn't require much expertise and is adequate for small businesses or non-critical applications.

A step up is to do load testing on the actual equipment before purchasing it or reassigning workloads. This gives fairly accurate information, but it is cumbersome and the results are only useful for that particular configuration and load level tested.

The better way is to go virtual using capacity planning modeling software. Instead of having to set up, configure and test the actual equipment, it can all be done in a simulation. This makes it easy to test a variety of equipment configurations and workloads to find at what point bottlenecks appear. It can also be used on the entire set of equipment and applications used to deliver a service, even if they are using different platforms, and so determine the final effect on service response times.

Although most economies are on an upswing and IT budgets are growing, according to the latest Gartner survey, it doesn’t mean that companies should return to the overprovisioning method. Instead, by using capacity modeling to establish just the right level of equipment, IT staff can continue to add business value, whether we are in a bull or a (three) bear market. That way, even if budgets are tight, life doesn’t have to be so Grimm.