40 years of searching for the best computer system response time

For over 40 years, system response time has been a topic of interest and controversy in computer science. Since the late 1960s, the field has seen numerous studies conducted and articles written addressing the issue. Many factors were measured in these studies including: users’ accuracy and error rates with different levels of system response time, user performance speed and the efficiency of the commands used, how user interactions with the computer changed as a result of changes in system response time, how their bodies reacted physiologically to those same changes and even how happy, satisfied, anxious or annoyed they were as system response times changed.In this paper, we summarize the major issues in system response time research and look at what can be concluded from them. Generally, researchers have suggested specific response-time guidelines based on the complexity of the task or the type of interaction with the computer. We suggest that rather than system response time being task- or expectation-focused, instead interactions with a computer fall into two categories: control tasks and conversational tasks. For control tasks, immediate response times are necessary for optimal user performance whereas for conversational tasks, some delays may be necessary to maintain the optimal pacing of the on-going conversation. The location and duration of these delays will depend on both task complexity and user expectations. Future system response time research is needed to further quantify limits of delay detection, and the location and duration of inter-task delays to optimize user performance and satisfaction with computers.

► We survey over 40 publications in the area of computer system response time (SRT).
► We show that consistent results are only found for low-level tasks.
► We show that previous models by Shneiderman and Seow have limited utility.
► We propose that control tasks and conversational tasks are fundamentally different for SRT.