The title of the paper -- Exodus Project - Pigs Really Do Fly! -- is a remark made by a staff member who, early on, thought the project was impossible.
The Unisys Clearpath Dorado mainframe was long dependable for the agency, and was running more than 2,000 programs. However, the model in use at the agency had reached its end of life and it had to be replaced by this year.
Albert said he would have stayed on this mainframe for a number of years if Unisys extended its life as well as cut MIPS (Millions of Instructions per Second) charges, which were running about $1 million a year. The agency paid an additional $200,000 in maintenance fees.
The MIPS charges and the end-of-life was grating for Albert.
He compared the situation to "owning a car that runs fine, but there is only one place you can buy gas for the car and that gas station refuses to sell any more gas for the car after that x-date."
"There was no reason why we couldn't use the mainframe another five years," said Albert. The latest machine had been installed in 2001 but was actively maintained it by adding fiber cards, I/O processors, I/O channels, as needed.
But there were other motivations beyond MIPS for migrating off the mainframe, namely its Pacbase code. Working with it was becoming increasingly difficult to support as staff retired. "There are very few Pacbase programmers in the world anymore," said Albert.
There was skepticism internally about whether the legacy code could work on a Windows platform and as reliably as it did on the mainframe. But with about 30% of the IT staff eligible for retirement in three years and 50% eligible in five years, a migration was all but inevitable.
The agency will still need Pacbase skills for years to come, but the long-term plan is to rewrite the two million lines of legacy code for a .Net environment. "As long as my Pacbase programmers stick around long enough to keep things running, my .Net programmers can rewrite it," he said.
Albert's paper outlines some of the struggles faced by the development team.
There "was no expert judgment available or models to follow," wrote Albert. The initial work could have been completed more quickly "had the decision to re-platform not been repeatedly debated."
Albert also made note of the "valleys of despair" the team faced along the way as well as including a list of lessons learned.
Despite its challenges, the team had executive sponsorship throughout the project.
Users were also engaged. "The users were more confident with the change because the re-platforming approach meant the same code was executing, and that meant very few errors occurred," wrote Albert.