![]() |
|
|
|
Systems sizing: Nobody gets it rightNo matter how sophisticated the sizing models, estimating hardware performance for enterprise resource planning systems is more art than science.By Larry Marion
Considering the extent of Rod Ely's pioneering efforts, he has surprisingly few arrows in his back. He is a systems architect at Green Mountain Coffee Roasters of Waterbury, Vt., and one of the first implementers of the manufacturing modules of the PeopleSoft ERP suite. Imagine installing a new enterprise resource planning (ERP) system in your shop and having to increase the clerical staff because of poor systems performance, and you get a sense of what Ely has faced over the past two years.
During sizing efforts, "we didn't realize how many batch jobs we would need to do," Ely explained to me after a presentation at a recent PeopleSoft user conference. "We just looked at online processing needs." The additional staffing at Green Mountain, along with database and applications fine-tuning and a new four-way Compaq server with Intel's new Xeon processors, will cure the performance ills. Ely is not alone, of course, when it comes to under-estimating the amount of processor horsepower needed to run ERP systems. The consensus of the scores of IT managers I've interviewed over the years, leads me to conclude that every ERP implementation team fails to accurately predict its hardware needs. Unfortunately for IT managers, there are as many reasons for the unpleasant performance surprises as there are servers in their datacenters. Knowing right from wrong Green Mountain endured the pain of being a pioneer--it's hard to determine the required processing power if you're among the first to use a package. Indeed, most organizations new to client/server ERP don't even understand the questions in the vendor sizing questionnaires, let alone know the correct answers. Other common trip wires on the road to accurate systems sizing are scope-creep and heightened user expectations of response times. Furthermore, occasionally you'll be victimized by aggressive vendors submitting low-ball bids in the hope of making their numbers from the follow-up orders. In addition, sometimes preliminary predictions of the reporting needs of individuals in an organization have no relation to financial reality--users want to produce and print hundreds of reports, without realizing the heavy performance penalty of report writing on production systems. Another pitfall is an ass-backwards approach to sizing. Scoping out a business case for the new systems is a critical first step. This should be done before software is selected and hardware studied, warns Jeanne Bauhart, vice president for enterprise resource solutions at Cambridge Technology Partners, a highly regarded implementation firm. "Most ERP buyers buy the software, then they estimate transactions and the number of bodies using it, then they figure the scope of the project (including features and functions), and then they say, 'Oh wow.' That's the wrong order." Here's the right way: After determining the business goals and objectives to be achieved, prioritize the capabilities, features, and functions under consideration, so your organization is in agreement on what it wants to accomplish before it sees the price tags and the feature/function trade-offs among the various software suites. After an organization has mapped packages and priorities and determined which package will best support its business goals, it's ready to look at hardware sizing models. Do your homework Leading hardware vendors and implementation partners will be quick to offer you the use of their sizing models. Be prepared for serious homework--these are not pop quizzes but major essay exams. And while they are useful exercises, they cannot be reliable predictors of your organization's needs and behaviors once an ERP system goes into production. Doing your own stress analysis is a popular option among large shops. Ford Motor earlier this year conducted a four-month evaluation of servers from Hewlett-Packard and Sun Microsystems. It built load-testing scripts and tried other performance prediction techniques. Ford then pushed the vendors as much as possible, since it expects thousands of users of its PeopleSoft human resources modules by the year 2001. Company sources say their stress testing showed that the HP boxes as configured by HP would not be able to handle the worst-case load, so Ford opted for Sun's 6500 line. Over the next few years the company will phase in six machines to provide fail-over protection.
| Home page | Cover story | Online ad index |
DATAMATION Copyright © 1995-1998. All rights reserved
|