Listen

Description

1994 to 1995: Sharing code and processes to build Office pays off and the concept of the suite takes hold in the market. The Windows schedule becomes a death march to finish what is hoped to be a revolutionary new operating system. Netscape IPOs in the summer of 1995, weeks before the public availability of Windows 95.

This is the first of five sections in this chapter on building Office 95.

Back to 034. Office 94, 96

Office94 had a plan in place, with a date and resources, as well as the major architectural bets of 32-bits and suite. There was so much focus on getting started and the resource constraints so extreme that, while this was the first release of an Office product, it would also be the last release for which the planning was fully distributed to each team with little centralized effort.

The organization was set with a feature team for each product and one in OPU. For the most part, the feature team leads were running this project, at least from the start, as the more senior leaders were supposed to focus more on Office96. Putting in place the plan for Office96 happened in parallel with a different set of people and a lot more attention from the BUMs. Unless there was something specifically troubling, for the most part Office94 was flying under the radar of DAD and had almost a skunkworks feel to it, at least at the start. That did not last long, for me.

As the new PM leader in Office, I was wearing two hats juggling both releases—in hindsight the first 12 months in Office were probably the time I put in the most hours working as I was learning how to manage managers for the first time, not to mention learning the DAD ways of working.

While I was juggling, among our team of about 15 program managers Mike Conte (MikeCon) was the primary program manager on OPU driving Office94. Mike joined the Excel team after experience running his own Macintosh startup. On Excel, he was a key to Excel gaining adoption at large customers, especially banks. A true New Yorker, Mike wore all black and had a no-nonsense attitude. He laid the groundwork for the release, but early on moved to a new role on the Windows team. He easily transitioned the release work to Heikki Kanerva (HeikkiK). Products had a formal management team that was ultimately accountable, but each release saw people step up and take on the informal leadership required for complex cross-group work. This was Heikki’s release to step up.

Heikki was a Finnish Olympic-caliber skier who found his way to Microsoft by studying business and computer science on a scholarship to the University of Alaska, after mandatory Finnish military service on a submarine. He joined DAD in the shared interoperability group and cut his teeth on shipping the enormously complex OLE technology. As it turned out, Heikki had absolutely the perfect demeanor to lead all of the teams through shipping the first synchronous release of Office. He was unflappable and possessed a military precision and the dedicated work ethic of an Olympian. And when needed, he could make light of a situation with a Finnish submariner expression that did not translate well at all into business English or, frankly, polite company.

Heikki led the newly formed Office-wide project meetings—the first time that key managers across all the apps assembled routinely to make progress on work. These meetings started off weekly and then as milestones approached, daily and then twice daily. While this was only a few people at the beginning, it was the organizational and cultural shift required to begin to operate as a suite, not as a bundle. These meetings became the operating heartbeat of the team, not a tax while the real work happened elsewhere.

I know it seems ridiculous to mention that we scheduled a weekly meeting for a new project, but up until this point creating Office essentially flew under the radar without an operational motion.

From an engineering perspective, the move to 32-bits was fairly smooth especially since much of the work had been done earlier in the side projects of making native apps for Windows NT as proofs of concept (and to make them available for sale). From a practical perspective, however, the transition to 32-bits hit Microsoft where it cared the most, performance. If there was one thing deep in the culture of the company, it was squeezing the most out of the scarce compute resources of CPU and memory. The move to 32-bits was inevitable, but the impact on performance was counter-intuitive.

In moving all the apps to 32-bits, no surprise but all the code got twice as big. And much of the data stored in memory got twice as big. Moving all that around made things slower. Crazy. For Office, 32-bits didn’t mean very much—Word already handled long documents, and spreadsheets could be pretty big. These products grew up in extremely limited memory.  In fact, many of our own tests were showing that the 16-bit versions of apps performed better on Windows 3.1 and were even a bit slower on Chicago. That’s because Chicago was also going through this widening experience with the operating system code. Everything was expanding. Except the system requirements, the amount of memory a PC required for Chicago and Office as printed on the box.

It didn’t take long for fingers to start pointing. That’s natural. The real issue was not necessarily the benchmarks which over time would tend to work themselves out (we hoped) but just how much memory was required to get reasonable performance. The key market promise for Chicago was that customers could upgrade their Windows 3 systems to Chicago and things would get better, but if those systems had only 4MB of memory or yikes 2MB then these systems would be horribly slow, and even more so if they also upgraded Office.

PCs were not only expensive in the 1990s, but they were also treated as capital expenses by companies. They had 5-year lifecycles, amortization schedules, and an expectation that the software could change over time without expensive and very difficult or impossible to implement hardware upgrades. Upgrading memory required a day or more of downtime, a “truck to roll” with a tech, and probably hundreds of dollars of hardware.

There was not much we could do about this but continue to work to reduce memory usage. There were some significant efforts in tools and analysis and some amazing work across Windows and Office to make things work on 4MB, but ultimately 8MB was “recommended” (versus “required”). This fell short of the 2MB Windows 3 required. As it would turn out, doubling the system requirements for each release of Windows became the norm as I would point out in about 12 years when we launched Windows 7. Office trended remarkably well and for many releases, the typical Office application (Word, Excel, PowerPoint) did not substantially increase requirements of about 2-4MB per application on top of the OS.



This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit hardcoresoftware.learningbyshipping.com