[ICFP'26] Deterministic Concurrency

Deterministic Concurrency (Video, ICFP 2026) Edward Lee (University of California at Berkeley) Abstract: It is not uncommon, particularly in the early literature in computer science, to equate concurrency with nondeterminism and to treat the two together as a profound difficulty encountered when trying to understand programs. This makes the title of this talk an oxymoron. Nondeterminism is often defined for programs as exhibiting “observably different behavior given the same inputs.” But nearly all programs exhibit observably different behavior given the same inputs. For example, the time at which an output is produced is observably different for all but the most stringent real-time systems. In practice, unless you are designing such a real-time system, you are unlikely to consider the timing of the output part of the “behavior.” What about the timing of the inputs? In this talk, I argue for interpreting determinism as a property of models, not as a property of physical systems. I will then argue that, contrary to almost all current practice, determinism has immense value in concurrent and distributed software systems. Moreover, I will show that deterministic models do not intrinsically come with a performance penalty and that they make the design of fault-tolerant systems easier, not harder. Presentation at the ICFP 2026 conference, August 25-27, 2026, https://icfp26.sigplan.org/ Sponsored by ACM SIGPLAN.