programming in the
twenty-first century

It's not about technology for its own sake. It's about being able to implement your ideas.

Progress Bars are Surprisingly Difficult

We've all seen progress bars that move slowly for twenty minutes, then rapidly fill up in the last 30 seconds. Or the reverse, where a once speedy bar takes 50% of the time covering the last few pixels. And bars that occasionally jump backward in time are not the rarity you'd expect them to be.

Even this past month, when I installed the macOS Sierra update, the process completed when the progress bar was only two-thirds full. DOOM 2016 has a circular progress meter for level loads, with the percent-complete in the center. It often sits for a while at 0%, gets stuck at 74% and 99%, and sometimes finishes in the 90s before reaching 100%.

Clearly this is not a trivial problem, or these quirks would be behind us.

Conceptually, a perfect progress bar is easy to build. All you need to know is exactly how long the total computation will take, then update the bar in its own thread so it animates smoothly. Simple! Why do developers have trouble with this? Again, all you need to know is exactly how long...

Oh.

You could time it with a stopwatch and use that value, but that assumes your system is the standard, and that other people won't have faster or slower processors, drives, or internet connections. You could run a little benchmark and adjust the timing based on that, but there are too many factors. You could refine the estimate mid-flight, but this is exactly the road that leads to the bar making sudden jumps into the past. It's all dancing around that you can't know ahead of time exactly how long it should take for the progress bar to go from empty to full.

There's a similar problem in process scheduling, where there are a number of programs to run sequentially in batch mode. One program at a time is selected to run to completion, then the next. If the goal is to have the lowest average time for programs being completed, then best criteria for choosing the next program to run is the one with the shortest execution time (see shortest job next). But this requires knowing how long each program will take before running it, and that's not possible in the general case.

And so the perfect progress bar is forever out of reach, but they're still useful, as established by Brad Allan Meyers in his 1985 paper ("The importance of percent-done progress indicators for computer-human interfaces"). But "percent-done" of what? It's easy to map the loading of a dozen similarly sized files to an overall percentage complete. Not so much when all kinds of downloading and local processing is combined together into a single progress number. At that point the progress bar loses all meaning except as an indication that there's some sort of movement toward a goal, and that mostly likely the application hasn't hasn't locked up.

(If you liked this, you might enjoy An Irrational Fear of Files on the Desktop.)

permalink December 21, 2016

previously

archives

twitter / mail

I'm James Hague, a recovering programmer who has been designing video games since the 1980s. Programming Without Being Obsessed With Programming and Organizational Skills Beat Algorithmic Wizardry are good starting points. For the older stuff, try the 2012 Retrospective.

Where are the comments?