April 12, 2004
(Computerworld)
After several years of making do with existing IT infrastructure, many companies are once again considering major purchases. But there's a huge difference between the buying process today and that of several years ago. "In the '90s, it was almost like having a free checkbook to just go out and buy what you need," says George Lewandowski, capacity planning supervisor at Metavante Corp., a Milwaukee-based company that sells IT products and services to financial institutions. "Organizations are finally starting to say that if they are going to spend money, they want to have a good ROI and TCO, not 50% utilization."
To achieve this, IT planners first must ensure that existing infrastructure elements are used to the fullest extent without lowering service levels. Then they must justify expenditures for new IT hardware by accurately predicting future needs.
Capacity planning processes and tools are essential for accurate planning. Lewandowski uses TeamQuest Corp.'s TeamQuest Model software to plan capacity needs for about 1,000 Windows and Unix servers. By using a mix of trending analysis and modeling to properly assign resources to tasks, he says he has driven server utilization up without affecting service-level agreements. "With proper capacity planning we can drive utilization into the 80% to 90% range and really start getting our money's worth out of the equipment," Lewandowski says. The tools available can produce very accurate results, he says, "provided you use them in a scientific manner."
Formula for Excess
Capacity planning is more than an application. Successful projects require a combination of skills, organizational processes and tools in order to ensure that a company has enough infrastructure to meet projected needs without overspending. Yet many planners take a more cautious route, buying excess capacity in order to ensure that there is never a shortage. "Most organizations use the formula CP = BMH: capacity planning equals 'buy more hardware,'" says Milind Govekar, an analyst at Gartner Inc. "True capacity planning is a process that requires a high level of IT maturity."
Image Credit: Jason Schneider
Almost any performance monitoring software can do this. But many packages that tout capacity planning capabilities don't do much more than track historical performance, and a performance trend isn't the best indicator of future requirements -- though in some cases it may be all you need, analysts say. "Trending ... doesn't work as well as most of its proponents imagine. However, it does enable reasonably accurate quick-and-dirty analyses to take place, with a good return on investment for this work," says Andy Bolton, CEO of Capacitas Ltd., a London-based capacity planning consultancy.
More advanced capacity planning software does more than track historical trends. It also lets IT planners create analytic models of different parts of the infrastructure to see how changes in hardware, applications or users will affect performance levels.
"If the project that is under way has a large impact, or large potential impact, it is essential to use modeling, since that will give you the best insight," says Glenn O'Donnell, an analyst at Meta Group Inc. in Stamford, Conn. But, he adds, "for minor projects, it is best to use trending, since it is too time-consuming to model everything."
Either way, analysts say the use of capacity planning tools is growing and taking on greater importance in IT operations. A recent Meta Group study of critical management issues put capacity management and planning at the top of the list for large enterprises.
But while the use of these tools is increasing, administrators aren't necessarily using them to predict future needs. Mike Ehr, research director at Enterprise Management Associates Inc. in Boulder, Colo., says organizations often use the tools to solve current problems connected with server consolidation and overprovisioning.
A Different Model
Capacity planning tools that perform modeling rather than just trending fall into three major categories. Mainframe-based tools come from traditional mainframe management vendors. Examples include Patrol Perform & Predict from Houston-based BMC Software Inc. and MXG from Merrill Consultants in Dallas.
In the server-based tools arena, the leading stand-alone vendor is Clear Lake, Iowa-based TeamQuest, but server management tools from vendors such as BMC and SAS Institute Inc. in Cary, N.C., also include server modeling features.
Network-based tools, such as those from Opnet Technologies Inc. in Bethesda, Md., focus on network planning. In addition, enterprise network management suites from Hewlett-Packard Co., Computer Associates International Inc. and IBM also include network capacity planning.
The tools in these categories overlap functionally, but generally each of the products performs best in its niche and offers lesser capabilities in the others. As a result, companies tend to use different tools for different parts of their infrastructures.
Boris Gdalevich, capacity planning manager at Quest Diagnostics Inc., a medical testing and diagnostics firm in Teterboro, N.J., uses BMC's Perform & Predict to model his servers while the network operations staff uses different tools. Sprint Corp. uses TeamQuest to model its midrange servers and PerfMan software from Bethlehem, Pa.-based The Information Systems Manager Inc. (ISM) to model the CPUs on its IBM z900 series mainframes.
Although using a single planning tool throughout the enterprise is desirable, analysts say it's best to use different tools for different areas because of the limitations of today's tools. Most companies have separate teams managing mainframes, servers and the network anyway, so each team should pick the tool that best suits its needs. Eventually, however, as companies move toward more dynamic, on-demand computing environments, everything will need to be modeled together.
"When enabled with automated provisioning and service-level management tools that detect changing needs and share the load on the fly, businesses should be able to reduce overall capital investment," says Mary Johnston Turner, vice president for enterprise strategies at Boston consulting firm Summit Strategies Inc. "I expect that in the future, capacity planning will be considered more a part of dynamic infrastructure provisioning and configuration/change management than as a stand-alone, application-specific activity."
Capacity planning tools are useful only if they produce accurate predictions. Fortunately, the tools are becoming more accurate. When Metavante ran one set of simulations using TeamQuest, the software predicted that the company would reach 80% utilization within two months. The actual number turned out to be 85%. The model also predicted that installing specified hardware would drop utilization to 65%. When Metavante put in that hardware and checked the actual utilization, the result was 66%. "We were very, very close," says Lewandowski.
Joel Allen, senior capacity planning manager at Sprint, says that the mainframe CPU models he has created using ISM's PerfMan have all yielded predictions that have come within 10% of the actual performance figures. Nevertheless, Allen says, "capacity planning is definitely more of an art than a science."
One reason is that accurate forecasts depend on getting people from different business units to give accurate and timely data on their future needs. Unlike trending, which makes predictions by reviewing historical performance data, modeling requires getting sales forecasts and other predictions from other parts of the organization and adding it to the model. It won't do the company any good to launch a big advertising campaign for its online services without having the bandwidth and server capacity to handle the projected increase in customers.
Jean-Pierre Garbani, an analyst at Forrester Research Inc. in Cambridge, Mass., agrees that capacity planning is "still a black art in many senses." No matter how sophisticated the analytics in the tools, the user still needs to have a thorough understanding of what parameters to model and then must interpret the data and ensure that it makes sense. Still, he doesn't advise trying to do sophisticated planning without using automated tools.
"If you are doing server consolidation or upgrading your operating system, even if you are very astute and skilled, you still need tools," he says. "Doing it by hand won't work."
Robb is a freelance writer based in Los Angeles. You can reach him at drewrobb@sbcglobal.net.