Chapter 2: The Product Without a Home

Mid-February through March 1995 — Novell offices and UnixWare engineering facilities, Utah

Paul’s fax came through on a Tuesday, twelve curling pages of thermal paper still warm when Sarah pulled them from the machine: three years of Front Range Commercial Bank’s server purchases, sorted the way Paul had sorted everything for fifteen years — by platform, then by date, a column of NetWare purchase orders running down most of every page.

The first page told Paul’s story exactly the way he wanted it told. NetWare remained overwhelmingly the platform of record. The bank kept buying NetWare servers on a schedule that hadn’t changed in three years. Nothing in the raw numbers suggested Front Range was going anywhere.

Sarah spent the morning taking Paul’s twelve pages apart and putting them back together by a different measure. Not total servers owned — new workload placed. She drew four columns on a legal pad: NetWare, commercial UNIX, other, and a fourth column she left unlabeled until she’d filled it in.

The existing file and print work kept landing on NetWare, exactly as it always had. The core banking applications and the databases they depended on kept landing on the UNIX systems, exactly as they always had. But three of the last six new departmental applications hadn’t landed on either. They’d arrived as complete packages, hardware and operating system and support contract bundled together by whichever vendor had won the department’s business that quarter. Two of those bundles ran on whatever platform their vendor happened to ship — a proprietary check-sorting controller nobody at the bank had chosen on its own merits, and a proprietary branch-automation system Front Range had never used before and would probably never use again. The third was the imaging pilot Ellen had shown her, and it was the first of the six to arrive on Windows NT. Three more NT purchases already sat in the current capital budget, waiting on nothing but department sign-off.

Sarah labeled the fourth column WINDOWS NT and looked at what she’d built. The installed base described where the bank had been. The purchase orders described where it was going. NetWare wasn’t losing a single server it already owned. It was losing every workload that hadn’t been invented yet — and there were always more of those than there were servers already sitting in a rack.

She moved the imaging pilot from the bottom of somebody’s mental NetWare tally to the top of a new list, and started making calls.

Margaret Sloan found her before lunch, arms full of the material Sarah had requested two weeks earlier — a research analyst’s idea of a complete answer, considerably more complete than Sarah had expected for a request she’d made half-formed from her Denver hotel room.

The financial documents explained, in the flat language finance people used when they’d already made up their minds, exactly why UnixWare kept losing every executive argument. The acquisition and the engineering investment behind it were difficult to justify against the revenue the product actually returned. UnixWare was competing for enterprise UNIX customers who already had relationships with Sun or HP or IBM, none of whom had any reason to switch. Novell’s own sales channel, built and trained to sell NetWare into network administrators, had no natural instinct for selling a general-purpose operating system to a UNIX buyer who had never thought of Novell as anything but a networking company. Supporting a full commercial UNIX — certification, driver work, a release engineering discipline separate from NetWare’s — cost money that showed up on a balance sheet long before the product justified it.

The technical material told a different story entirely, and Sarah read it twice to be sure she wasn’t only seeing what she wanted to see. UnixWare did the things NetWare had never been built to do and never would be: a genuine multiuser environment, System V interfaces every UNIX developer already knew, a real home for the databases and line-of-business applications that had outgrown a specialized network kernel, and all of it running on Intel hardware that cost a fraction of what a Sun or HP box charged for the same workload. UnixWare 2, barely a month old, was current and competent by any technical measure Sarah could find.

Novell had a financial argument for giving up on UnixWare and a technical argument for keeping it, and the two arguments were not describing the same product. A poorly positioned product was not automatically a poor operating system. Somebody in this building had confused the two, and it was costing Novell an asset nobody else in the industry could easily replace.

She went to see the people who actually built it.

Eric Langford met her in the UnixWare engineering lab already braced for a fight, which told Sarah most of what she needed to know about how these visits usually went. He was younger than she’d pictured from his memos, and he had the specific tiredness of an engineer who had spent two years defending work that kept getting reviewed by people deciding whether to sell it.