“The key problem in collaboration is not communication — it is concurrency.”
The Story
In 1969, Clarence "Skip" Ellis became the first African American to earn a Ph.D. in computer science, at the University of Illinois at Urbana-Champaign. The year is worth sitting with — 1969 was the year Apollo 11 landed on the Moon. The computing field was not yet twenty years old as a formal discipline. Ellis was not just among the first to do advanced research in the field; he was there while the field itself was being invented. He had already worked at Bell Labs as a teenager — hired in the late 1950s to operate a computer, he taught himself to program it before anyone thought to offer him formal training.
He went to Xerox PARC.
PARC in the late 1970s is one of the most celebrated research environments in the history of computing — the place where the graphical user interface, Ethernet, laser printing, and object-oriented programming all either originated or were significantly advanced. Ellis was head of the Office Research Group, and his team built OfficeTalk: an office automation system that used icons in a graphical interface to represent files and workflows, connected workstations across Ethernet, integrated email, and supported collaborative work among groups of users simultaneously. OfficeTalk is rarely mentioned in the same breath as the Alto or Smalltalk. But it was doing something the others weren't: it was thinking about computers as a medium for collaboration, not just computation. Single-user productivity was the dominant assumption of the era. Ellis's group was asking what happened when you networked people, not just machines.
He left PARC and went to the Microelectronics and Computer Technology Corporation (MCC). In 1989, working with Simon Gibbs, he published the paper that introduced Operational Transformation — the algorithm that would, thirty years later, run inside Google Docs.
The problem OT solves is the hardest problem in collaborative editing: what happens when two users edit the same document at the same time, from different machines, with network latency between them? If Alice inserts a character at position 5, and Bob deletes a character at position 3, and both operations happen simultaneously — in what order should they be applied? With naive synchronization, you get either conflicts or corruption. OT solved this by transforming each operation to account for the operations that happened concurrently. It preserved the intent of each edit, regardless of the order in which they arrived.
This is a concurrency problem — the same class of problem documented in this museum's Temporal Coupling pattern. But where most concurrency solutions use locks — preventing simultaneous access — OT embraces concurrency and resolves conflicts mathematically after the fact. It is lock-free collaborative editing, and it works.
The algorithm is non-trivial. The paper, "Concurrency Control in Groupware Systems," is mathematically dense. It took fifteen years for the industry to build reliable implementations. Google Wave was a failed attempt. Google Docs eventually got it right. Every real-time collaboration tool you have used in the last decade — Google Docs, Figma (which uses a descendant called CRDT), Notion, VS Code Live Share, Overleaf — runs on some form of the concurrency model that Ellis and Gibbs formalized. The seventeen-year gap between a foundational algorithm and its first successful mass-market application is the kind of thing this museum documents. The pattern was there. Implementation was waiting for conditions to catch up.
Ellis spent the later part of his career at UC Boulder, where he continued researching collaborative systems and mentoring students. He passed away in 2014.
Why They're in the Hall
Ellis is a Pioneer and Builder whose core contribution is the antidote to Temporal Coupling — one of this museum's six root failure mechanics.
Most exhibits in the Temporal Coupling pattern class document what goes wrong when concurrent operations collide: race conditions, data corruption, the Unsynchronized Handshake. Ellis's work demonstrates that concurrency is not inherently dangerous — it is dangerous when the system doesn't have a mathematical model for resolving conflicts. Operational Transformation provides that model. It doesn't prevent concurrent edits. It makes them safe by transforming them into a convergent sequence.
Ellis did this work at Xerox PARC — the same institution where Alan Kay's team created Smalltalk and the GUI, where Bob Metcalfe invented Ethernet, where the modern computer was essentially designed. He was one of the few Black researchers at PARC during its golden era. His contribution — making concurrent collaboration mathematically sound — is used by billions of people daily, none of whom know his name.
The pattern Ellis solved is the same pattern that crashes database systems, corrupts files, and causes the race conditions documented throughout this museum. He solved it for human collaboration. The rest of the industry is still working on solving it for everything else.