System Administrators are Users Too

Rob Barrett, IBM Almaden Research Center
barrett@almaden.ibm.com

Seminar on People, Computers, and Design
Stanford University May 30, 2003

Most human-computer interaction work has focused on end users of computing systems. Another important class of computer users, however, is the cohort of administrators who design, build, maintain and troubleshoot computer systems. These highly-expert users are vital for the operation of our "e-everything" world, yet little effort has gone into studying their work and developing tools that help them be effective. This is especially important because the labor associated with operating large computational systems is increasingly outstripping the cost of the technology itself.

Our research group is performing a series of ethnographic studies of system administrators in their work environments. This presentation will include results from these studies, as well as information developed at a CHI2003 workshop on system administration as users; this workshop brought together researchers, developers, and practitioners from industria and academia.

From this group and from our own work, a consistent set of paradoxes is beginning to emerge. First, tremendous effort has gone into the design of powerful GUI tools for system administration. Many tools have been developed and validated with established user-centered design methodologies. Yet field studies repeatedly find system administrators ignoring these tools and falling back on the standard command shell and least-common denominator tools such as 'grep' and 'vi'. Second, system administration is a highly collaborative activity, with a heavy dependence on instant messaging, email, telephone, and face-to-face interaction. Yet system administration tools rarely include collaboration aids, instead seemingly assuming that these workers toil away silently and alone. Third, effective operation and problem resolution requires an accurate mental model of how the system functions. "Situation awareness" theory dictates that a model starts with sensory input, develops with mental comprehension, and results in predictions of system behavior. Yet large-scale systems have few and unintegrated sensing mechanisms, and are too complex for any single person to comprehend, resulting in unpredictable behavior.

This presentation will illustrate each of the three paradoxes with examples from field experience, and offer suggestions for how the HCI community can move forward to resolve them.

Rob Barrett is a Research Staff Member at the IBM Almaden Research Center in California where he works in the Services Research group that aims to bring value from HCI to the IBM Global Services organization. His current work focuses on the user experience of system administration and human aspects of autonomic computing. Previous work includes an intermediary approach to designing web applications, optimization of pointing devices, track-following servo systems for tape data storage, and atomic-scale imaging. He holds a Ph.D. in Applied Physics from Stanford University and has earned masters and bachelors degrees in physics, electrical engineering and theology. He has over 40 refereed publications and 16 patents in fields ranging from applied math to physics and computer science..

View this talk on line at CS547 on Stanford OnLine

Titles and abstracts for all years are available by year and by speaker.

For more information about HCI at Stanford see