The Machine Was Busy

Batch computing made the separation between action and consequence institutional. A programmer prepared a job, delivered it into an operational system, and received its printed outcome later. Fernando Corbató and his colleagues described the contrast when presenting an experimental time-sharing system in 1962: under batch-monitor conditions, the correction of even one mistake could require hours or an entire day. The processor might discover the mistake immediately. The human delay was assembled around that instant by collection, scheduling, operation, printing and return.

That distinction matters because machine time and experienced time are rarely the same quantity. The IBM 650 is remembered partly for bringing data processing into more continuous organisational use, yet its existence inside an institution still required people, media and schedules to coordinate around scarce equipment. A calculation can be fast while access to calculation is slow. The earliest waiting room in this museum is therefore not inside a processor. It is the office, loading dock and operator desk surrounding it.

Personal machines moved computation into view, but they did not make it transparent. Storage retained physical motion. A floppy drive placed a head above a track and waited for a rotating sector; a cassette carried every earlier second past a fixed head before reaching later data. These mechanisms leaked evidence. Rhythmic stepping suggested a search. A steady tape tone suggested continuity. Repeated movement could mean recovery, while a sudden silence could feel ominous. Users learned to treat incidental sound as operational telemetry.

This was not an interface designed in the modern sense. The sound was a consequence of motors, heads and signals, yet it acquired meaning through repetition. People listened because the screen offered less information than the machine body. The result was a distributed display: one part visual, one part acoustic, one part remembered from prior failures. When solid-state devices removed the mechanical soundtrack, computers became quieter and, in a narrow sense, less legible. Designers then had to manufacture signs that old hardware emitted for free.

Brad Myers tested percent-done indicators with 48 participants in research published in 1985. The study reported that 86.1 percent preferred the version that displayed progress, with a reported probability of p = .0006. The result is modest and important. A bar did not accelerate the task, improve its answer or return control to the participant. It changed the experience of a fixed delay by replacing an unbounded question with a bounded one: this much has happened, and some amount remains.

Yet the bar introduces another obligation. To show a proportion, software needs a credible numerator and denominator. Many operations discover their workload only by doing it. Others contain phases that cannot be compared by counting. The visual confidence of a smooth line can therefore exceed the confidence of the underlying estimate. A good indicator is not a window onto the processor. It is an editorial decision about which uncertainty to expose, which to compress, and when to admit that total duration is unknown.

Rendering demonstrates the persistence of busyness after computation becomes massively parallel. Pixar described a production environment in which, at the peak of Finding Nemo, roughly 3,000 processors still faced more than 10,000 queued tasks. The figures are not a universal measure of animation production, but they expose a durable rule: added capacity often invites added ambition. More frames, richer light and greater revision consume the speed that engineering creates.

At that scale, nobody listens to a particular processor. Evidence moves upward into dashboards, frame lists, error summaries and scheduling policies. The individual machine disappears into an institution, much as it did in the batch room. The modern phrase “the system is busy” may refer to a fleet rather than a box, but the trust problem is the same. A person needs to know whether work exists, whether it is advancing, and whether waiting remains the rational choice. Interfaces have become quieter; their responsibility to account for time has not.