Transition tour outlines future

HP unveils cost estimates, range of tools in road show

Second of three parts

Birket Foster is looking for some help at HP’s Houston, Texas stop of its Transition tour. On the one-year anniversary of HP’s decision to pull out of the 3000 market, this date in the 12-city road show includes a segment on planning for the spending that moves HP 3000 customers onto other systems. In the meeting room at the Houston Hilton, Foster wants someone to help add up the numbers around migration’s costs.

One customer carries a calculator and volunteers to do the math. Then the attendee — one of about 50 in the room including customers, HP partners, job-seekers and consultants — discovers his calculator isn’t working. Even without the grand totals at hand, Foster’s numbers are eye-openers for customers.

Foster’s segment illustrates the work that customers are most likely to do in their migration efforts during 2003: planning to spend money. A very small portion of the 3000 customer base had budgeted for any expense in moving applications off their HP 3000; the system’s cost of ownership was one of its chief attractions. When HP announced an end date for its support and sales, a new category was forced onto IT budgets.

In the Houston hotel room Foster is comfortable with his presentation, even though the numbers elicit little response from attendees. It’s like being told they need a life-saving operation, but one which will cost more than they have to spend — for today, anyway.

Those companies choosing to turn off their HP 3000s will need new money to do the job. “You’ll get an idea of how much to go ask your management for,” Foster promises. The numbers have been worked up across all of the North American Platinum migration partners, but in Houston Foster gets to deliver the hard news — hard in part because it starts with the costs of new hardware.

HP’s Platinum partners state the costs in ranges, and on this day in Houston the bottom of the range is a free hardware option. Converting A-Class and N-Class hardware to HP-UX hardware units is free with HP’s conversion kits. All other hardware choices run from $15,000 for a low-end system, from $50,000 for a midrange-sized system, and starting at $100,000 on up to $1 million in the medium range. The slides show that the highest end hardware costs start at $1 million for HP’s Unix solutions.

On the next slide the cost of hardware replacement gets the NT alternative, a relief compared to the Unix numbers. HP’s figures show anywhere from $2,000 to $8,500 per server for an NT replacement. But Foster reminds users that Windows server licensing can get expensive, and to budget for a total expense between $10,000 to $20,000 if following the NT route. That figure that doesn’t include storage, high availability configurations or a database. The NT solution could still run as high as $1 million for hardware.

Linux hardware numbers look the same as NT solutions in HP’s cost estimates, but the cost of the operating system can be much lower. HP offers a secure version of Linux that costs about $3,000.

Alternative hardware expenses, while lower than the HP 3000’s prices, don’t include databases. Once again there’s mention of a free option, but the PostgreSQL and MySQL databases only run under Linux. The IMAGE work-alike Eloquence costs $7,000, while better-known alternative Oracle can be had at a discount of $20,000 per processor. Microsoft’s SQL Server runs in-between at $10,000-$20,000 per server.

The tour’s slides also quote prices between $3,000 and $23,000 per server for Informix. One slide notes that the database now owned by IBM “may not be a great strategic option,” a reference to a declining share of market that haunts the choice of Eloquence as well.

Foster’s slides show other hard costs: license transfer fees for 4GLs like Speedware and PowerHouse, where customers should expect $10,000-$200,000 per server; costs of replacing the HP COBOL II compiler with products like AcuCOBOL ($2,500 per developer, $150 for the first user, $23 per runtime user, by HP’s accounting), Microfocus ($3,000 per developer and $187 per runtime user), or Fujitsu’s NetCOBOL ($3,000 per developer, no runtime fees.)

The slides provide no cost estimates for replacing tools like spoolers, backup products, job schedulers and database managers. But migration costs for labor and consulting are estimated at $100,000-$250,000 to outsource Speedware migrations, for example, or $100,000 to move 1 million lines of COBOL code.

Migrating databases to those Oracle and SQL Server options will cost between $10,000 and $80,000 for a tool, and up to $200,000 for full migration, mirroring and load testing tools. Finally, there’s no estimate offered for training of programmers, operations staff, hiring database administrators (see sidebar) or re-training end users on new applications.

“It takes a huge amount of training,” Foster tells the crowd, which has remained quiet during the dozens of slides detailing hundreds of thousands of dollars in expenses. In about 45 minutes, Foster has returned to the point that customers need to know what to do in their migration.

“From an end-user point of view, operations point of view, and programming point of view, people have to understand how the new system hangs together,” he says. “That’s a learning curve.”

Foster reminds the customers in the meeting room that much of their 3000 company training got done “in tribal learning,” where each person trains the next, and only 80 percent of the knowledge is passed on. “You get down to less than 50 percent of what you need to know to manage the application,” he says, “pretty scary. You’ll have to have a formal training program for your packaged applications.”

At the close of the hour, no final roll-up total is presented to the crowd. But using the bottom-end costs for every category — hardware and databases, compiler licensing, tools, and migration resources — would deliver a budget of more than $200,000. The rock-bottom figure assumes a customer will convert an HP 3000 to an HP-UX server at no cost, and have very few applications and a small database to migrate.

Making replacements

Foster holds forth on the realities of buying off-the-shelf software to replace home-grown applications. “You’re going to lose functionality along the way, and you may have over-buy a package to get the functionality you need,” he says. Some of MB Foster’s customers are on “a continuous treadmill, chasing [features in] applications as they come out with new versions.”

Even if a site buys a package, a customer may still need to migrate some applications as well. As an example Foster mentions interfaces to warehouse or EDI systems, which “you still may want to move to the new [packaged] system.” New functionality in any packaged application will change how a company does its day-to-day business, he adds.

Tools to move code

Nearing the lunch break, the attendees get their first crash course in migration software for the 3000. The programs are sold and recommended by all Platinum partners in the room, but Foster asserts that much more than software is required for a successful migration. “A fool with a tool is still a fool,” he says.

The potential for failures in migrations are well-documented during this segment. A pie chart from the Standish Group notes that among ERP implementations the consulting group surveyed, more than one third failed and were cancelled. More than half of the projects had cost overruns that averaged 178 percent of planned expenditures. And those projects which made it to completion didn’t achieve much more than 60 percent of the planned functionality.

There’s more than one approach to migration software, too. Moving applications with tools like Neartek’s AMXW and Denkart’s ViaNova 3000 differs from Transoft’s offerings. “The first two have an MPE emulation layer,” Foster says. “Transoft takes you into the native layer.”

The length of that journey apparently varies widely. Foster offers notes on timelines for migration projects, but the estimates vary widely. COBOL-VPlus migration time runs from six months to five years, depending on the complexity of application code. Database migration takes between one to three months, while Transact and PowerHouse migrations are estimated between 6 months and four years. Only Speedware application migrations show a finish date under one year, with an estimate of three to nine months.

Beyond lunch, a full plate

The attendees break away from the meeting room to gather modest white-box lunches: cold-cut sandwiches and small bags of chips with iced cans of soda and some fruit and cookies. The crowd mingles in the hall outside the meeting room for the first time since getting the cost details on moving off the 3000. A few attendees network to look for IT jobs or consulting contracts.

After the meal, Speedware’s Chris Koppe takes on the specifics of migrating COBOL and Fortran applications, including a brief review of the available third party COBOL compilers for HP-UX. Little is said on this day about Linux’s compilers — or much else about the Open Source alternative — since so much of the presentation is tuned for HP’s Unix alternatives.

Koppe notes that “pretty much nobody here has a plan” to migrate yet, observing that few hands that go up when he asks for a progress report from customers. The attendance has dropped from the morning session, though much of the attrition comes from the ranks of consultants and HP partners. Customers hang in, listening to Koppe talk about their applications, which the customers describe as not modular. Instead they’ve got applications laced with VPlus and intrinsic calls, the kind Koppe says are “much more difficult” to migrate.

“You may need to consider emulation as an alternative for this kind of application,” he says. The solutions such as Denkart’s and Neartek’s support nearly all the configurations in current MPE applications. “You’d be pretty hard-pressed to find options they wouldn’t support,” Koppe says of the emulation products’ VPlus compatibility. “They’d have to be options that weren’t commonly used.”

But these emulation choices come with residual, yearly support fees after purchasing the tool from the supplier. Neartek’s solution “is a product you purchase and do it yourself,” according to Koppe, while Denkart’s ViaNova is a service-based offering which takes code and returns it migrated to the target environment. Neartek also has service partners who can perform migrations using AMXW. Both solutions will note which code couldn’t be migrated in error reports. Customers can then contract for the extra consulting to finish the migration, or do the last bits themselves.

The talk covers a trio of COBOL alternatives for HP-UX, from AcuCOBOL to Microfocus COBOL to Fujitsu’s NetCOBOL. Koppe warns of reserved words in some of these solutions, like WINDOW in AcuCOBOL, for example — a valid data item name in the 3000’s COBOL II.

Koppe runs through options for the balance of potential languages on the HP 3000: Fortran, Pascal, RPG, C, SPL and even Business Basic, noting alternatives for each language. It’s a great example of the richness of information promised by the tour. The presentation even notes things like the fact that Pascal will be supported under HP-UX, but not in native Itanium mode. Koppe even mentions that Pascal is coming off HP’s HP-UX price list in two years.

RPG programs — maybe the rarest of HP 3000 languages — can be converted by a tool from Richter Software. Attendees hear that C programs can be moved by Denkart’s ViaNova 3000 service; SPL code, in far greater use than either RPG or C, can be moved with a SPLash compiler forthcoming from Allegro Software and AD Technologies. The new version of SPLash moves the code to HP-UX, Windows or Linux. ViaNova and AMXW also migrate SPL to C under those solutions’ MPE emulation. Even Business Basic gets covered. Marxmeier Software offers Business Basic migration through its Eloquence database, which includes Business Basic.

Solutions for VPlus migration and emulation get reviewed, briefly. Denkart’s EdWin is a graphical version of VPlus emulation for non-MPE platforms, as is App2XML from Cheops. ScreenJet has a migration toolkit to convert VPlus to AcuCOBOL’s Acubench, and AD Technologies has a migration toolkit which also targets Acubench.

With the daunting array of technology in review, Koppe suggests that the customers in the room think of the Platinum partners as “general contractors.” Speedware, MB Foster, MBS and Lund Performance Solutions are in the room in Houston to show their expertise as experts managing relationships between sites and the unfamiliar new technical resources.

Koppe, like Foster before him, wants to stress how much the migration effort will tax customers. “Don’t under-estimate how many resources it take,” he says, “especially for testing. Getting it converted is one thing, getting it tested is another.”

Next issue: How refactoring can deliver more than a migration, and how customers react to the briefing.