Demystifying capacity planning

By his own admission, John Payton used to plan network capacity for R.R. Donnelley & Sons Co. by the seat of his pants.

“I knew how I did things in the past, and I figured if it worked once, it’s worth trying it again,” Payton says. “If we came up short, I’d just throw more bandwidth on the network. It really wasn’t a scientific method.”

But when the Chicago publisher changed its business model and expanded its services to include content management and digital photography, Payton realized even his best “educated guess” might not be enough to ensure the network could handle new bandwidth-hogging applications.

Payton is not alone. Historically, network architects would simply add more bandwidth to networks to improve performance when their businesses added more users and services. But Payton points out the art of capacity planning “isn’t as simple as ‘more is better.'”

There are many capacity planning tools on the market that collect network, system and application performance data – in real time or over long periods, or both. Some software lets companies create model networks and simulate traffic to predict their bandwidth needs for future users and applications. The modelling tools let companies pose “what if?” scenarios to calculate how the network would behave in certain situations.

About two years ago, Payton and his team at R.R. Donnelley started using Opnet Technologies Inc.’s software called IT Guru. He tested how new applications would behave on his network and determined exactly how much bandwidth he’d need. In the past, he would have made an estimate and doubled that number, but the software helps him determine more accurately what to provision.

Payton uses the software to analyse the response times of applications in the simulated network and determine how he could improve them, whether through server configuration or consolidation, adding bandwidth or rewriting an application.

Danny Gornell, chief architect at online brokerage firm Ameritrade Holding Corp., tells a similar story about building out his company’s network when the economy boomed a few years ago. Gornell says he and his staff had to rebuild Ameritrade’s infrastructure every nine months because their techniques for capacity planning fell short when predicting rapid growth.

“Our planning was done completely anecdotally. It was a seat-of-the-pants exercise that really lacked any analytical data,” Gornell says.

Gornell and his staff started using BMC Software Inc.’s two offerings that work together for capacity planning, Patrol Perform and Patrol Predict. He could determine the number of Web servers needed to support the expanding online business, without overprovisioning the network.

Mostly, BMC’s software helped Gornell determine the number of servers he’d need to keep delivering business services, and the software predicts the impact of system changes before they occur. BMC software agents reside on each managed server and report application response time, CPU utilization and network availability. The capacity planning software then uses the data to project what will be needed to sustain future business services and users.

Now with the economy in the doldrums, Gornell says his days of overprovisioning have ended, and he’s using the same tools from BMC to uncover underutilized resources and avoid purchasing more hardware. “Now we’re using it in a diametrically opposed way. We’re trying to consolidate our systems and servers, and still meet business objectives,” he says.

Although Payton and Gornell say the tools have helped them better plan their networks, the two are quick to add that these software tools do not represent a cure-all for capacity planning woes. Capacity planning requires knowledge of how all the network elements interact and perform. That only comes from years of study and experience with networks.

“This is not a plug-and-play tool. It requires the right human resources and human interaction to do the correct analysis. It’s not going to do the work of the engineers,” Gornell says.

Payton likens his capacity planning tool to a Magic 8 Ball: “If you ask it the wrong question, it will still give you a nifty answer, but that may not solve your problem. You need to know the right questions to ask.”

He says capacity planning is an art form – one that requires an artist skilled in advanced mathematics and statistical analysis, with an eye for spotting behaviour patterns in volumes of data.

Industry experts echo those warnings and go a step further, explaining why the current capacity planning offerings will not solve many of a network executive’s problems. Users have not expressed much interest in long-term planning tools, opting more for quick fixes that sometimes don’t solve the problems and prompt network managers to overprovision their nets, says Jean-Pierre Garbani, an analyst at Giga Information Group Inc.

Garbani says vendors will better address capacity planning in 2002. For now, he says only BMC comes close to helping network architects account for all the elements in a distributed environment when planning future nets. And although he says BMC has the capability to offer a complete capacity planning tool kit, its current products don’t address all users’ needs, partly because demand for such tools is low.

Capacity planning tools from vendors such as Computer Associates International Inc. and Hewlett-Packard Co. are fundamentally performance management tools with some extra reporting capabilities, Garbani says. Others such as Opnet and Compuware offer the modelling and predictive analysis he thinks is essential to capacity planning in large enterprise networks – but they look only at a single function, such as server capacity, application performance or WAN bandwidth.

Jeff Nudler, an analyst at Enterprise Management Associates Inc., agrees there are no tools currently available that completely address capacity planning needs. He says network architects need to know in real time how to plan for capacity in their networks.

Clairvoyant Inc.’s ForeCast Manager – software that collects real-time and historical data, spotting behaviour patterns and offering fixes – most closely fits the bill for Nudler, but even this tool can only predict as well as the user can.

As Payton and Gornell say, the tools do not make decisions or do the planning. The tools expedite the collection of data from many disparate sources of an infrastructure and offer some correlation. But in terms of predicting the network future, the network architect in part controls the software’s outcome.

“Capacity planning has long been considered the black magic of the IT industry. How does one predict the future?” Nudler says. “Even with all the tools, equations, algorithms and mathematics, it’s impossible to calculate how a network will behave.”